In this project we investigated the electrophysiological correlates of the combined processing of basic visual properties (i.e., size and contrast) and the emotional content of simple words.
GRAPHS
grand.avg.localizer <-
read_csv(here::here("EEG/main/data/main_grandEEG_localizer.csv")) %>%
select(-condition) %>% # drop condition column
gather( # convert to long format
key = timepoint, # name of new column
value = amplitude, # value of new column
-c(participant, electrode) # keep participants and electrodes as columns
)
# summarized data from each time point (& within-subject 95% CI)
grand.avg.localizer.pointsummary <-
grand.avg.localizer %>%
summarySEwithin(
data = .,
measurevar = "amplitude",
withinvars = c("electrode", "timepoint"),
idvar = "participant"
) %>%
mutate(
timepoint = as.numeric(levels(timepoint))[timepoint] # re-convert time points to numeric
)
# plot
ggplot(
grand.avg.localizer.pointsummary,
aes(
x = timepoint,
y = amplitude,
group = electrode
)
) +
geom_vline( # vertical reference line
xintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_hline( # horizontal reference line
yintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_vline( # vertical reference lines
xintercept = seq(-200, 1000, 100),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_hline( # horizontal reference lines
yintercept = seq(-3, 3, 1),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_line( # one line per electrode
size = 1.2,
color = "#494847",
alpha = .8
) +
geom_ribbon( # 95% CI
aes(
ymin = amplitude - ci,
ymax = amplitude + ci
),
linetype = "dotted",
size = .1,
alpha = .1,
show.legend = FALSE
) +
labs(
title = "", # title & axes labels
x = "time (ms)",
y = expression(paste("amplitude (", mu, "V)"))
) +
scale_x_continuous(breaks = seq(-200, 1000, 100)) + # x-axis: tick marks
scale_y_reverse(
breaks = seq(-3, 3, 1), # y-axis: tick marks
limits = c(3, -3)
) +
theme_classic(base_size = 20) +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
legend.position = "none"
)
Butterfly plot
Prepare data for:
- topographies
- waveforms
- (RDI plots) pirateplots
# load electrode locations
electrodeLocs <-
read_delim(here::here("EEG/main/data/electrodes/elecLocs_BioSemi64.locs"),
"\t",
col_names = c("chanNo", "theta", "radius", "electrode"),
escape_double = FALSE, trim_ws = TRUE
) %>%
mutate(
radianTheta = pi / 180 * theta, # convert theta values from degrees to radians
# Cartesian coordinates
x = radius * sin(radianTheta),
y = radius * cos(radianTheta)
) %>%
.[order(.$electrode, decreasing = FALSE), ] # re-order according to topos
# load topographies
topos <-
read_csv(here::here("EEG/main/data/main_topoEEG.csv")) %>%
group_by(bin, electrode) %>% # average across participants
dplyr::summarize(
P1 = mean(P1),
N1 = mean(N1),
EPN = mean(EPN),
LPP = mean(LPP)
)grand.avg.comps <-
read_csv(here::here("EEG/main/data/main_avgERP.csv")) %>%
gather( # convert to long format
key = timepoint,
value = amplitude,
-c(participant, elec_cluster, condition)
) %>%
# rename levels
mutate(
condition = revalue(
as.factor(condition),
c(
"all" = "localizer",
"negative" = "negative",
"negative minus neutral" = "negative minus neutral",
"negativeLargeBright" = "negative large bright",
"negativeLargeDark" = "negative large dark",
"negativeSmallBright" = "negative small bright",
"negativeSmallDark" = "negative small dark",
"neutral" = "neutral",
"neutralLargeBright" = "neutral large bright",
"neutralLargeDark" = "neutral large dark",
"neutralSmallBright" = "neutral small bright",
"neutralSmallDark" = "neutral small dark"
)
),
# variable with main effect of size
size = revalue(
condition,
c(
"localizer" = NA,
"negative" = NA,
"negative minus neutral" = NA,
"negative large bright" = "large",
"negative large dark" = "large",
"negative small bright" = "small",
"negative small dark" = "small",
"neutral" = NA,
"neutral large bright" = "large",
"neutral large dark" = "large",
"neutral small bright" = "small",
"neutral small dark" = "small"
)
),
# variable with main effect of contrast
cont = revalue(
condition,
c(
"localizer" = NA,
"negative" = NA,
"negative minus neutral" = NA,
"negative large bright" = "bright",
"negative large dark" = "dark",
"negative small bright" = "bright",
"negative small dark" = "dark",
"neutral" = NA,
"neutral large bright" = "bright",
"neutral large dark" = "dark",
"neutral small bright" = "bright",
"neutral small dark" = "dark"
)
),
# variable with main effect of emotion
emo = revalue(
condition,
c(
"localizer" = NA,
"negative" = NA,
"negative minus neutral" = NA,
"negative large bright" = "negative",
"negative large dark" = "negative",
"negative small bright" = "negative",
"negative small dark" = "negative",
"neutral" = NA,
"neutral large bright" = "neutral",
"neutral large dark" = "neutral",
"neutral small bright" = "neutral",
"neutral small dark" = "neutral"
)
)
)# trial-wise words presented to each participant
data.EEG.trialWords <- read_csv(here::here("EEG/main/data/main_words.csv"))
data.EEG.trial <-
read_csv(here::here("EEG/main/data/main_trialEEG.csv")) %>%
mutate(
participant = as.factor(participant),
condition = factor(condition),
# variable with main effect of size
size = revalue(
condition, c(
"1" = "large",
"2" = "small",
"3" = "large",
"4" = "small",
"5" = "large",
"6" = "small",
"7" = "large",
"8" = "small"
)
),
# variable with main effect of contrast
cont = revalue(
condition, c(
"1" = "dark",
"2" = "dark",
"3" = "bright",
"4" = "bright",
"5" = "dark",
"6" = "dark",
"7" = "bright",
"8" = "bright"
)
),
# variable with main effect of emotion
emo = revalue(
condition, c(
"1" = "negative",
"2" = "negative",
"3" = "negative",
"4" = "negative",
"5" = "neutral",
"6" = "neutral",
"7" = "neutral",
"8" = "neutral"
)
),
cond = paste(emo, size, cont, sep = " ") # rename condition levels
) %>%
bind_cols(., data.EEG.trialWords) %>% # include words in EEG data frame
dplyr::select( # re-order variables
participant,
word,
size,
cont,
emo,
cond,
P1,
N1,
EPN,
LPP
)P1
topos %>%
dplyr::filter(bin == "localizer") %>%
dplyr::select(electrode, amplitude = "P1") %>%
bind_cols(., electrodeLocs) %>% # bind electrode locations and amplitudes (necessary)
topoplot(.,
limits = c(-1, 1.5), # min-max amplitude
chanLocs = electrodeLocs,
method = "Biharmonic",
palette = "viridis",
interp_limit = "skirt",
contour = TRUE,
chan_marker = "point",
quantity = "amplitude",
highlights = c("P7", "P9", "PO7", "O1", "O2", "PO8", "P8", "P10")
) +
ggtitle("P1 (localizer)") +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
plot.margin = unit(c(6, 0, 6, 0), "pt") # decrease plot margins
)
Amplitude extracted from the following electrode cluster: P7 P9 PO7 O1 O2 PO8 P8 P10.
# summarized data from each time point, including within-subject 95% CIs
P1.grand.avg.comps.pointsummary <-
grand.avg.comps %>% # data frame
filter( # filter out unused conditions
elec_cluster == "P1",
condition != "negative",
condition != "negative minus neutral",
condition != "neutral"
) %>%
summarySEwithin(
data = .,
measurevar = "amplitude",
withinvars = c("size", "cont", "emo", "condition", "timepoint"),
idvar = "participant"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>% # for graph
mutate(
visualfeats = as.factor(ifelse(visualfeats == "NA NA", "localizer", visualfeats)),
timepoint = as.numeric(levels(timepoint))[timepoint]
)
# plot waveforms
ggplot(
data = filter(
P1.grand.avg.comps.pointsummary,
condition != "localizer"
),
aes(
x = timepoint,
y = amplitude,
linetype = emo,
color = visualfeats
)
) +
geom_vline( # vertical reference line
xintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_hline( # horizontal reference line
yintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_vline( # vertical reference lines
xintercept = seq(-200, 1000, 50),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_hline( # horizontal reference lines
yintercept = seq(-3, 3, 1),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_line( # one line per condition
size = 2.5,
alpha = .6
) +
geom_line( # thick black line (localizer)
data = filter(
P1.grand.avg.comps.pointsummary,
visualfeats == "localizer"
),
aes(
x = timepoint,
y = amplitude
),
size = 5,
show.legend = FALSE,
inherit.aes = FALSE
) +
geom_ribbon( # localizer 95% CI
data = filter(
P1.grand.avg.comps.pointsummary,
visualfeats == "localizer"
),
aes(
ymin = amplitude - ci,
ymax = amplitude + ci
),
linetype = "dotted",
size = .1,
alpha = .3,
show.legend = FALSE
) +
scale_color_manual(
breaks = c(
"localizer", "large bright", # re-order conditions
"large dark", "small bright", "small dark"
),
values = waveforms.palette
) + # line colors
scale_fill_manual(values = waveforms.palette) + # ribbon colors
labs(
title = "P1 (66-148 ms)", # title & axes labels
x = "time (ms)",
y = expression(paste("amplitude (", mu, "V)"))
) +
scale_x_continuous(
breaks = seq(-200, 1000, 50), # x-axis: tick marks
limits = c(-100, 250)
) +
scale_y_reverse(
breaks = seq(-3, 3, 1), # y-axis: tick marks
limits = c(3, -3)
) +
annotate("rect", # highlight time window used for analysis
xmin = 66,
xmax = 148,
ymin = -.5,
ymax = 3,
linetype = "solid",
size = 2,
color = "#de1d1d",
alpha = 0
) +
theme_classic(base_size = 20) +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
legend.position = "none"
)
summarySEwithin(
data.EEG.trial,
"P1",
withinvars = c("participant", "size", "cont", "emo"),
idvar = "word"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>% # for graph
pirateplot(
formula = P1 ~ emo + visualfeats, # dependent ~ independent variables
data = ., # data frame
main = "P1", # main title
xlim = NULL, # x-axis: limits
xlab = "", # x-axis: label
ylim = c(-4, 6), # y-axis: limits
ylab = expression(paste("amplitude (", mu, "V)")), # y-axis: label
inf.method = "hdi", # type of inference: 95% Bayesian Highest Density Interval (HDI)
inf.within = "participant", # ID variable in within-subject designs
hdi.iter = 5000, # number of iterations for 95% HDI
cap.beans = TRUE, # bean densities capped at data limits
point.cex = 1.3, # point size
pal = pirateplot.palette # color palette
)
N1
topos %>%
dplyr::filter(bin == "localizer") %>%
dplyr::select(electrode, amplitude = "N1") %>%
bind_cols(., electrodeLocs) %>%
topoplot(.,
limits = c(-1, 1),
chanLocs = electrodeLocs,
method = "Biharmonic",
palette = "viridis",
interp_limit = "skirt",
contour = TRUE,
chan_marker = "point",
quantity = "amplitude",
highlights = c("TP7", "P7", "P9", "TP8", "P8", "P10")
) +
ggtitle("N1 (localizer)") +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
plot.margin = unit(c(6, 0, 6, 0), "pt") # decrease plot margins
)
Amplitude extracted from the following electrode cluster: TP7 P7 P9 TP8 P8 P10.
N1.grand.avg.comps.pointsummary <-
grand.avg.comps %>%
filter(
elec_cluster == "N1",
condition != "negative",
condition != "negative minus neutral",
condition != "neutral"
) %>%
summarySEwithin(
data = .,
measurevar = "amplitude",
withinvars = c("size", "cont", "emo", "condition", "timepoint"),
idvar = "participant"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>%
mutate(
visualfeats = as.factor(ifelse(visualfeats == "NA NA", "localizer", visualfeats)),
timepoint = as.numeric(levels(timepoint))[timepoint]
)
# plot waveforms
ggplot(
data = filter(
N1.grand.avg.comps.pointsummary,
condition != "localizer"
),
aes(
x = timepoint,
y = amplitude,
linetype = emo,
color = visualfeats
)
) +
geom_vline(
xintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_hline(
yintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_vline(
xintercept = seq(-200, 1000, 50),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_hline(
yintercept = seq(-3, 3, 1),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_line(
size = 2.5,
alpha = .6
) +
geom_line(
data = filter(
N1.grand.avg.comps.pointsummary,
visualfeats == "localizer"
),
aes(
x = timepoint,
y = amplitude
),
size = 5,
show.legend = FALSE,
inherit.aes = FALSE
) +
geom_ribbon(
data = filter(
N1.grand.avg.comps.pointsummary,
visualfeats == "localizer"
),
aes(
ymin = amplitude - ci,
ymax = amplitude + ci
),
linetype = "solid",
size = .1,
alpha = .3,
show.legend = FALSE
) +
scale_color_manual(
breaks = c(
"localizer", "large bright",
"large dark", "small bright", "small dark"
),
values = waveforms.palette
) +
scale_fill_manual(values = waveforms.palette) +
labs(
title = "N1 (150-260 ms)",
x = "time (ms)",
y = expression(paste("amplitude (", mu, "V)"))
) +
scale_x_continuous(
breaks = seq(-200, 1000, 50),
limits = c(-100, 400)
) +
scale_y_reverse(
breaks = seq(-3, 3, 1),
limits = c(3, -3)
) +
annotate("rect",
xmin = 150,
xmax = 260,
ymin = -2.5,
ymax = 1,
linetype = "solid",
size = 2,
color = "#de1d1d",
alpha = 0
) +
theme_classic(base_size = 20) +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
legend.position = "none"
)
summarySEwithin(
data.EEG.trial,
"N1",
withinvars = c("participant", "size", "cont", "emo"),
idvar = "word"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>%
pirateplot(
formula = N1 ~ emo + visualfeats,
data = .,
main = "N1",
xlim = NULL,
xlab = "",
ylim = c(-7, 3),
ylab = expression(paste("amplitude (", mu, "V)")),
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
EPN
topos %>%
dplyr::filter(bin == "localizer") %>%
dplyr::select(electrode, amplitude = "EPN") %>%
bind_cols(., electrodeLocs) %>%
topoplot(.,
limits = c(-1, 1), # min-max amplitude
chanLocs = electrodeLocs,
method = "Biharmonic",
palette = "viridis",
interp_limit = "skirt",
contour = TRUE,
chan_marker = "point",
quantity = "amplitude"
) +
ggtitle("EPN (localizer)") +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
plot.margin = unit(c(6, 0, 6, 0), "pt") # decrease plot margins
)
Because the topography differs from the classical EPN topography published in the literature, we identified this component by calculating the difference between negative and neutral conditions (averaged across size and contrast). This is in accordance with our pre-registered protocol.
Note that this was not necessary when identifying the LPP.
topos %>%
dplyr::filter(bin == "NegMinNeut") %>%
dplyr::select(electrode, amplitude = "EPN") %>%
bind_cols(., electrodeLocs) %>%
topoplot(.,
limits = c(-.3, .4), # min-max amplitude
chanLocs = electrodeLocs,
method = "Biharmonic",
palette = "viridis",
interp_limit = "skirt",
contour = TRUE,
chan_marker = "point",
quantity = "amplitude",
highlights = c("P7", "P9", "PO7", "PO3", "O1", "Oz", "Iz", "O2", "PO4", "PO8", "P8", "P10")
) +
ggtitle("EPN (negative minus neutral)") +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
plot.margin = unit(c(6, 0, 6, 0), "pt") # decrease plot margins
)
Amplitude extracted from the following electrode cluster: P7 P9 PO7 PO3 O1 Oz Iz O2 PO4 PO8 P8 P10.
EPN.grand.avg.comps.pointsummary <-
grand.avg.comps %>%
filter(
elec_cluster == "EPN",
condition == "negative minus neutral" |
condition == "negative large bright" |
condition == "negative large dark" |
condition == "negative small bright" |
condition == "negative small dark" |
condition == "neutral large bright" |
condition == "neutral large dark" |
condition == "neutral small bright" |
condition == "neutral small dark"
) %>%
summarySEwithin(
data = .,
measurevar = "amplitude",
withinvars = c("size", "cont", "emo", "condition", "timepoint"),
idvar = "participant"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>% # for graph
mutate(
visualfeats = as.factor(ifelse(visualfeats == "NA NA", "negative minus neutral", visualfeats)),
timepoint = as.numeric(levels(timepoint))[timepoint]
)
# plot waveforms
ggplot(
data = EPN.grand.avg.comps.pointsummary,
aes(
x = timepoint,
y = amplitude,
linetype = emo,
color = visualfeats
)
) +
geom_vline(
xintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_hline(
yintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_vline(
xintercept = seq(-200, 1000, 50),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_hline(
yintercept = seq(-3, 3, 1),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_line(
size = 2.5,
alpha = .6
) +
geom_line(
data = filter(
EPN.grand.avg.comps.pointsummary,
visualfeats == "negative minus neutral"
),
aes(
x = timepoint,
y = amplitude
),
size = 5,
show.legend = FALSE,
inherit.aes = FALSE
) +
geom_ribbon(
data = filter(
EPN.grand.avg.comps.pointsummary,
visualfeats == "negative minus neutral"
),
aes(
ymin = amplitude - ci,
ymax = amplitude + ci
),
linetype = "dotted",
size = .1,
alpha = .3,
show.legend = FALSE
) +
scale_color_manual(
breaks = c(
"negative minus neutral", "large bright",
"large dark", "small bright", "small dark"
),
values = waveforms.palette
) +
scale_fill_manual(values = waveforms.palette) +
labs(
title = "EPN (300-500 ms)",
x = "time (ms)",
y = expression(paste("amplitude (", mu, "V)"))
) +
scale_x_continuous(
breaks = seq(-200, 1000, 50),
limits = c(-100, 600)
) +
scale_y_reverse(
breaks = seq(-2, 3, 1),
limits = c(3, -2)
) +
annotate("rect",
xmin = 300,
xmax = 500,
ymin = -1,
ymax = 2.5,
linetype = "solid",
size = 2,
color = "#de1d1d",
alpha = 0
) +
theme_classic(base_size = 20) +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
legend.position = "none"
)
summarySEwithin(
data.EEG.trial,
"EPN",
withinvars = c("participant", "size", "cont", "emo"),
idvar = "word"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>%
pirateplot(
formula = EPN ~ emo + visualfeats,
data = .,
main = "EPN",
xlim = NULL,
xlab = "",
ylim = c(-5, 5),
ylab = expression(paste("amplitude (", mu, "V)")),
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
LPP
topos %>%
dplyr::filter(bin == "localizer") %>%
dplyr::select(electrode, amplitude = "LPP") %>%
bind_cols(., electrodeLocs) %>%
topoplot(.,
limits = c(-1.5, 1.5), # min-max amplitude
chanLocs = electrodeLocs,
method = "Biharmonic",
palette = "viridis",
interp_limit = "skirt",
contour = TRUE,
chan_marker = "point",
quantity = "amplitude",
highlights = c("P1", "Pz", "P2", "P4", "P6", "P8", "P10", "POz", "PO4", "PO8")
) +
ggtitle("LPP (localizer)") +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
plot.margin = unit(c(6, 0, 6, 0), "pt") # decrease plot margins
)
Amplitude extracted from the following electrode cluster: P1 Pz P2 P4 P6 P8 P10 POz PO4 PO8.
LPP.grand.avg.comps.pointsummary <-
grand.avg.comps %>%
filter(
elec_cluster == "LPP",
condition == "localizer" |
condition == "negative large bright" |
condition == "negative large dark" |
condition == "negative small bright" |
condition == "negative small dark" |
condition == "neutral large bright" |
condition == "neutral large dark" |
condition == "neutral small bright" |
condition == "neutral small dark"
) %>%
summarySEwithin(
data = .,
measurevar = "amplitude",
withinvars = c("size", "cont", "emo", "condition", "timepoint"),
idvar = "participant"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>% # for graph
mutate(
visualfeats = as.factor(ifelse(visualfeats == "NA NA", "localizer", visualfeats)),
timepoint = as.numeric(levels(timepoint))[timepoint]
)
# plot waveforms
ggplot(
data = LPP.grand.avg.comps.pointsummary,
aes(
x = timepoint,
y = amplitude,
linetype = emo,
color = visualfeats
)
) +
geom_vline(
xintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_hline(
yintercept = 0,
linetype = "dashed",
color = "black",
size = 1.2,
alpha = .8
) +
geom_vline(
xintercept = seq(-200, 1000, 100),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_hline(
yintercept = seq(-3, 3, 1),
linetype = "dotted",
color = "#999999",
size = .8,
alpha = .5
) +
geom_line(
size = 2.5,
alpha = .6
) +
geom_line(
data = filter(
LPP.grand.avg.comps.pointsummary,
visualfeats == "localizer"
),
aes(
x = timepoint,
y = amplitude
),
size = 5,
show.legend = FALSE,
inherit.aes = FALSE
) +
geom_ribbon(
data = filter(
LPP.grand.avg.comps.pointsummary,
visualfeats == "localizer"
),
aes(
ymin = amplitude - ci,
ymax = amplitude + ci
),
linetype = "dotted",
size = .1,
alpha = .3,
show.legend = FALSE
) +
scale_color_manual(
breaks = c(
"localizer", "large bright",
"large dark", "small bright", "small dark"
),
values = waveforms.palette
) +
scale_fill_manual(values = waveforms.palette) +
labs(
title = "LPP (402-684 ms)",
x = "time (ms)",
y = expression(paste("amplitude (", mu, "V)"))
) +
scale_x_continuous(
breaks = seq(-200, 1000, 100),
limits = c(-100, 800)
) +
scale_y_reverse(
breaks = seq(-2, 3, 1),
limits = c(3, -2)
) +
annotate("rect",
xmin = 402,
xmax = 684,
ymin = -.5,
ymax = 2.5,
linetype = "solid",
size = 2,
color = "#de1d1d",
alpha = 0
) +
theme_classic(base_size = 20) +
theme(
plot.title = element_text(
size = 28,
hjust = .5,
face = "bold"
),
legend.position = "none"
)
summarySEwithin(
data.EEG.trial,
"LPP",
withinvars = c("participant", "size", "cont", "emo"),
idvar = "word"
) %>%
unite(visualfeats, c("size", "cont"), sep = " ") %>%
pirateplot(
formula = LPP ~ emo + visualfeats,
data = .,
main = "LPP",
xlim = NULL,
xlab = "",
ylim = c(-2, 4),
ylab = expression(paste("amplitude (", mu, "V)")),
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
CONFIRMATORY ANALYSIS
Calculate and compare the Bayes Factor of different linear mixed-effects models. Participants’ ID is included as random factor and its variance set as nuisance.
Compare (against the null model) the following models:
- main effects of size and emotion
- interactive effects of size and emotion
- main effects of contrast and emotion
- interactive effects of contrast and emotion
- main effects of size, contrast, and emotion
- interactive effects of size, contrast, and emotion
Afterwards, compare the best competing models to understand which one should be preferred overall.
P1
summarySEwithin(
data.EEG.trial,
"P1",
withinvars = c("size", "cont", "emo"),
idvar = "participant"
) %>%
kable(., digits = 2)| size | cont | emo | N | P1 | sd | se | ci |
|---|---|---|---|---|---|---|---|
| large | dark | negative | 2374 | 1.00 | 3.82 | 0.08 | 0.15 |
| large | dark | neutral | 2369 | 1.06 | 3.76 | 0.08 | 0.15 |
| large | bright | negative | 2373 | 1.21 | 3.76 | 0.08 | 0.15 |
| large | bright | neutral | 2364 | 1.34 | 3.76 | 0.08 | 0.15 |
| small | dark | negative | 2383 | 1.09 | 3.81 | 0.08 | 0.15 |
| small | dark | neutral | 2376 | 0.92 | 3.73 | 0.08 | 0.15 |
| small | bright | negative | 2365 | 0.34 | 3.83 | 0.08 | 0.15 |
| small | bright | neutral | 2363 | 0.38 | 3.86 | 0.08 | 0.16 |
for (k in 1:length(scaling.factor)) { # loop through scaling factors
### main effects of size and emotion
P1.sizeplusemo.BF <- lmBF(
P1 ~ size + emo + participant, # formula
data.EEG.trial, # data
whichRandom = "participant", # random effects
rscaleFixed = scaling.factor[k], # scaling factor of prior on effect size
rscaleRandom = "nuisance", # prior scale for standardized random effects
rscaleCont = "medium", # prior scale for standardized slopes
method = "simple", # MCMC sampling
iterations = niter # number of MCMC iterations
)
saveRDS(P1.sizeplusemo.BF, # save model
file = here::here(paste0("EEG/main/models/P1_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
### interactive effects of size and emotion
P1.sizebyemo.BF <- lmBF(
P1 ~ size * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/P1_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
### main effects of contrast and emotion
P1.contplusemo.BF <- lmBF(
P1 ~ cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.contplusemo.BF,
file = here::here(paste0("EEG/main/models/P1_contplusemo_BF_", scaling.factor[k], ".rds"))
)
### interactive effects of contrast and emotion
P1.contbyemo.BF <- lmBF(
P1 ~ cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.contbyemo.BF,
file = here::here(paste0("EEG/main/models/P1_contbyemo_BF_", scaling.factor[k], ".rds"))
)
### main effects of size, contrast, and emotion
P1.sizepluscontplusemo.BF <- lmBF(
P1 ~ size + cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/P1_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
### interactive effects of size, contrast, and emotion
P1.sizebycontbyemo.BF <- lmBF(
P1 ~ size * cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/P1_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
}# preallocate matrix with all BFs
compare.P1.BF <- matrix(NA, 6, length(scaling.factor))
# preallocate matrix with all % errors
compare.P1.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) { # loop through scaling factors
### load models
P1.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
P1.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_contplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_contbyemo_BF_", scaling.factor[k], ".rds"))
)
P1.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
# BFs
compare.P1.BF[, k] <- c(
P1.sizeplusemo.BF@bayesFactor$bf,
P1.sizebyemo.BF@bayesFactor$bf,
P1.contplusemo.BF@bayesFactor$bf,
P1.contbyemo.BF@bayesFactor$bf,
P1.sizepluscontplusemo.BF@bayesFactor$bf,
P1.sizebycontbyemo.BF@bayesFactor$bf
)
# percentage of error
compare.P1.perc.err[, k] <- c(
P1.sizeplusemo.BF@bayesFactor$error * 100,
P1.sizebyemo.BF@bayesFactor$error * 100,
P1.contplusemo.BF@bayesFactor$error * 100,
P1.contbyemo.BF@bayesFactor$error * 100,
P1.sizepluscontplusemo.BF@bayesFactor$error * 100,
P1.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
# summary confirmatory analysis
compare.P1 <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.P1.BF[, 1], "nar.p.err" = compare.P1.perc.err[, 1],
"med" = compare.P1.BF[, 2], "med.p.err" = compare.P1.perc.err[, 2],
"wid" = compare.P1.BF[, 3], "wid.p.err" = compare.P1.perc.err[, 3]
)
# sort according to medium scaling factor (in descending order)
compare.P1 <- compare.P1[order(compare.P1$med, decreasing = TRUE), ]
# nicer table
kable(compare.P1)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 6 | size x cont x emo | 516.892 | 16.2436 | 514.126 | 18.3575 | 512.777 | 39.6341 |
| 5 | size + cont + emo | 491.855 | 13.1956 | 490.608 | 10.9562 | 489.588 | 13.3064 |
| 1 | size + emo | 487.969 | 11.3261 | 487.267 | 10.8352 | 486.329 | 10.6004 |
| 2 | size x emo | 485.359 | 13.1905 | 484.269 | 12.3227 | 483.004 | 12.4829 |
| 3 | contr + emo | 455.029 | 11.1600 | 454.463 | 10.1229 | 453.767 | 10.9300 |
| 4 | cont x emo | 452.174 | 20.2480 | 451.206 | 14.2266 | 449.945 | 11.2667 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size x cont x emo ought to be preferred.
The best model (size x cont x emo) explains the observed data 1.6369e+10 times better than the second best model (size + cont + emo).
Paired comparisons
# interactions of interest (i.e., always negative vs. neutral)
P1.posthocBF <-
data.EEG.trial %>%
select(participant, cond, P1) %>%
summarySEwithin(data = ., measurevar = "P1", withinvars = c("participant", "cond")) %>%
split(.$cond) # split according to condition
for (k in 1:length(scaling.factor)) { # loop through scaling factors
### large, dark, negative vs. neutral
P1.posthocBF.large.dark.negVSneut <- ttestBF(
P1.posthocBF$`negative large dark`$P1, # first condition
P1.posthocBF$`neutral large dark`$P1, # second condition
mu = 0, # null hypothesis (mean difference = 0)
paired = TRUE, # paired sample test
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/P1_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
### small, dark, negative vs. neutral
P1.posthocBF.small.dark.negVSneut <- ttestBF(
P1.posthocBF$`negative small dark`$P1,
P1.posthocBF$`neutral small dark`$P1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/P1_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
### large, bright, negative vs. neutral
P1.posthocBF.large.bright.negVSneut <- ttestBF(
P1.posthocBF$`negative large bright`$P1,
P1.posthocBF$`neutral large bright`$P1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/P1_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### small, bright, negative vs. neutral
P1.posthocBF.small.bright.negVSneut <- ttestBF(
P1.posthocBF$`negative small bright`$P1,
P1.posthocBF$`neutral small bright`$P1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/P1_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
}# preallocate matrix with all BF10
compare.P1.posthocBF <- matrix(NA, 4, length(scaling.factor))
# preallocate matrix with all % errors
compare.P1.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
P1.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
P1.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
P1.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
P1.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### model comparison
compare.P1.posthocBF[, k] <- c(
P1.posthocBF.large.dark.negVSneut@bayesFactor$bf,
P1.posthocBF.small.dark.negVSneut@bayesFactor$bf,
P1.posthocBF.large.bright.negVSneut@bayesFactor$bf,
P1.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
# percentage of error
compare.P1.posthocBF.perc.err[, k] <- c(
P1.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
P1.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
P1.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
P1.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
# summary
compare.P1.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.P1.posthocBF[, 1], "nar.p.err" = compare.P1.posthocBF.perc.err[, 1],
"med" = compare.P1.posthocBF[, 2], "med.p.err" = compare.P1.posthocBF.perc.err[, 2],
"wid" = compare.P1.posthocBF[, 3], "wid.p.err" = compare.P1.posthocBF.perc.err[, 3]
)
kable(compare.P1.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.242476 | 0.000005 | -1.536763 | 0.000008 | -1.852970 | 0.000009 |
| small.dark.negVSneut | -0.083050 | 0.000004 | -0.305802 | 0.000002 | -0.575117 | 0.000001 |
| large.bright.negVSneut | -0.797664 | 0.000005 | -1.063337 | 0.000005 | -1.361115 | 0.000004 |
| small.bright.negVSneut | -1.410897 | 0.000005 | -1.716419 | 0.000010 | -2.039751 | 0.000011 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in P1 amplitude:
- large size, high contrast: 0.215076;
- small size, high contrast: 0.736532;
- large size, low contrast: 0.345302;
- small size, low contrast: 0.179709.
Exploratory model selection
### How important is each effect/interaction if we remove it from the full model?
for (k in 1:length(scaling.factor)) { # loop through scaling factors
P1.anovaBF <- anovaBF(
P1 ~ size * cont * emo + participant,
data.EEG.trial,
whichModels = "top", # test all models created by removing a main effect or interaction from the full model
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(P1.anovaBF,
file = here::here(paste0("EEG/main/models/P1_anovaBF_", scaling.factor[k], ".rds"))
)
}# preallocate matrix with all BF10
compare.P1.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
# preallocate matrix with all % errors
compare.P1.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load model
P1.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/P1_anovaBF_", scaling.factor[k], ".rds"))
)
# BFs
compare.P1.anovaBF.temp[, k] <- P1.anovaBF@bayesFactor$bf
# percentage of error
compare.P1.anovaBF.perc.err.temp[, k] <- P1.anovaBF@bayesFactor$error * 100
}
# summary exploratory analysis
compare.P1.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.P1.anovaBF.temp[, 1], "nar.p.err" = compare.P1.anovaBF.perc.err.temp[, 1],
"med" = compare.P1.anovaBF.temp[, 2], "med.p.err" = compare.P1.anovaBF.perc.err.temp[, 2],
"wid" = compare.P1.anovaBF.temp[, 3], "wid.p.err" = compare.P1.anovaBF.perc.err.temp[, 3]
)
compare.P1.anovaBF <- compare.P1.anovaBF[order(compare.P1.anovaBF$med, decreasing = TRUE), ]
kable(compare.P1.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | emo | 4.48535 | 39.1146 | 4.30235 | 57.3429 | 4.78873 | 34.6213 |
| 2 | cont:emo | 2.70675 | 25.6105 | 3.54899 | 54.3118 | 3.38254 | 35.0968 |
| 1 | cont:emo:size | 3.32238 | 26.0123 | 3.07197 | 38.4583 | 3.98027 | 38.4676 |
| 3 | emo:size | 2.77062 | 25.2758 | 2.47795 | 35.8667 | 3.17187 | 31.3004 |
| 6 | cont | -3.45884 | 37.1962 | -3.99157 | 35.7692 | -3.54226 | 31.1723 |
| 4 | cont:size | -33.87948 | 27.1131 | -34.55479 | 35.2074 | -33.70824 | 30.3864 |
| 7 | size | -36.90773 | 28.7470 | -36.66896 | 43.3900 | -36.52127 | 30.2185 |
Exploratory analyses suggest that omitting the factor emotion from the full model improves fitting by 73.87328 times. Removing the interactions size x contrast x emotion, contrast x emotion, and emotion x size also improves fitting by 34.778268, 21.584342, and 11.91681 times, respectively.
Conversely, omitting the factor contrast or the contrast x size interaction lowers the explanatory value of the resulting model by 54.139968 and 1.01614e+15 times. Finally, omitting the factor size is maximally detrimental, as it would lower the explanatory value of the resulting model by 8.41645e+15 times.
N1
summarySEwithin(
data.EEG.trial,
"N1",
withinvars = c("size", "cont", "emo"),
idvar = "participant"
) %>%
kable(., digits = 2)| size | cont | emo | N | N1 | sd | se | ci |
|---|---|---|---|---|---|---|---|
| large | dark | negative | 2374 | -1.36 | 3.67 | 0.08 | 0.15 |
| large | dark | neutral | 2369 | -1.38 | 3.69 | 0.08 | 0.15 |
| large | bright | negative | 2373 | -1.03 | 3.69 | 0.08 | 0.15 |
| large | bright | neutral | 2364 | -1.13 | 3.76 | 0.08 | 0.15 |
| small | dark | negative | 2383 | -1.39 | 3.67 | 0.08 | 0.15 |
| small | dark | neutral | 2376 | -1.45 | 3.71 | 0.08 | 0.15 |
| small | bright | negative | 2365 | -0.03 | 3.73 | 0.08 | 0.15 |
| small | bright | neutral | 2363 | -0.09 | 3.86 | 0.08 | 0.16 |
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
N1.sizeplusemo.BF <- lmBF(
N1 ~ size + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.sizeplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
### interactive effects of size and emotion
N1.sizebyemo.BF <- lmBF(
N1 ~ size * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/N1_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
### main effects of contrast and emotion
N1.contplusemo.BF <- lmBF(
N1 ~ cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.contplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_contplusemo_BF_", scaling.factor[k], ".rds"))
)
### interactive effects of contrast and emotion
N1.contbyemo.BF <- lmBF(
N1 ~ cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.contbyemo.BF,
file = here::here(paste0("EEG/main/models/N1_contbyemo_BF_", scaling.factor[k], ".rds"))
)
### main effects of size, contrast, and emotion
N1.sizepluscontplusemo.BF <- lmBF(
N1 ~ size + cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
### interactive effects of size, contrast, and emotion
N1.sizebycontbyemo.BF <- lmBF(
N1 ~ size * cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/N1_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
}compare.N1.BF <- matrix(NA, 6, length(scaling.factor))
compare.N1.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) { # loop through scaling factors
### load models
N1.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
N1.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_contplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_contbyemo_BF_", scaling.factor[k], ".rds"))
)
N1.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
# BFs
compare.N1.BF[, k] <- c(
N1.sizeplusemo.BF@bayesFactor$bf,
N1.sizebyemo.BF@bayesFactor$bf,
N1.contplusemo.BF@bayesFactor$bf,
N1.contbyemo.BF@bayesFactor$bf,
N1.sizepluscontplusemo.BF@bayesFactor$bf,
N1.sizebycontbyemo.BF@bayesFactor$bf
)
# percentage of error
compare.N1.perc.err[, k] <- c(
N1.sizeplusemo.BF@bayesFactor$error * 100,
N1.sizebyemo.BF@bayesFactor$error * 100,
N1.contplusemo.BF@bayesFactor$error * 100,
N1.contbyemo.BF@bayesFactor$error * 100,
N1.sizepluscontplusemo.BF@bayesFactor$error * 100,
N1.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
# summary confirmatory analysis
compare.N1 <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.N1.BF[, 1], "nar.p.err" = compare.N1.perc.err[, 1],
"med" = compare.N1.BF[, 2], "med.p.err" = compare.N1.perc.err[, 2],
"wid" = compare.N1.BF[, 3], "wid.p.err" = compare.N1.perc.err[, 3]
)
compare.N1 <- compare.N1[order(compare.N1$med, decreasing = TRUE), ]
kable(compare.N1)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 6 | size x cont x emo | 1136.051 | 17.90141 | 1133.523 | 8.82070 | 1131.017 | 8.16727 |
| 5 | size + cont + emo | 1095.365 | 3.72406 | 1094.384 | 4.01831 | 1093.388 | 3.95813 |
| 3 | contr + emo | 1054.595 | 3.24399 | 1053.978 | 3.33639 | 1053.309 | 3.20027 |
| 4 | cont x emo | 1050.861 | 3.72420 | 1049.929 | 4.09289 | 1048.905 | 3.97891 |
| 1 | size + emo | 968.706 | 3.41983 | 968.014 | 3.46315 | 967.314 | 3.39620 |
| 2 | size x emo | 964.924 | 4.17994 | 963.909 | 4.63483 | 962.822 | 4.64631 |
# BFs TOO BIG FOR R! WHEN "Inf", IT MEANS THAT THE RESULT IS > 1.8e+308!
# THAT'S WHY WE SWITCHED TO LOG_BF01When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size x cont x emo ought to be preferred.
The best model (size x cont x emo) explains the observed data 9.95963e+16 times better than the second best model (size + cont + emo).
Paired comparisons
N1.posthocBF <-
data.EEG.trial %>%
select(participant, cond, N1) %>%
summarySEwithin(data = ., measurevar = "N1", withinvars = c("participant", "cond")) %>%
split(.$cond)
### uncomment to recompute models ###
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
N1.posthocBF.large.dark.negVSneut <- ttestBF(
N1.posthocBF$`negative large dark`$N1,
N1.posthocBF$`neutral large dark`$N1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/N1_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
### small, dark, negative vs. neutral
N1.posthocBF.small.dark.negVSneut <- ttestBF(
N1.posthocBF$`negative small dark`$N1,
N1.posthocBF$`neutral small dark`$N1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/N1_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
### large, bright, negative vs. neutral
N1.posthocBF.large.bright.negVSneut <- ttestBF(
N1.posthocBF$`negative large bright`$N1,
N1.posthocBF$`neutral large bright`$N1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/N1_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### small, bright, negative vs. neutral
N1.posthocBF.small.bright.negVSneut <- ttestBF(
N1.posthocBF$`negative small bright`$N1,
N1.posthocBF$`neutral small bright`$N1,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/N1_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
}compare.N1.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.N1.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
N1.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
N1.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
N1.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
N1.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
compare.N1.posthocBF[, k] <- c(
N1.posthocBF.large.dark.negVSneut@bayesFactor$bf,
N1.posthocBF.small.dark.negVSneut@bayesFactor$bf,
N1.posthocBF.large.bright.negVSneut@bayesFactor$bf,
N1.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
compare.N1.posthocBF.perc.err[, k] <- c(
N1.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
N1.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
N1.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
N1.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
compare.N1.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.N1.posthocBF[, 1], "nar.p.err" = compare.N1.posthocBF.perc.err[, 1],
"med" = compare.N1.posthocBF[, 2], "med.p.err" = compare.N1.posthocBF.perc.err[, 2],
"wid" = compare.N1.posthocBF[, 3], "wid.p.err" = compare.N1.posthocBF.perc.err[, 3]
)
kable(compare.N1.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.44404 | 0.000005 | -1.75180 | 0.000010 | -2.07655 | 0.000012 |
| small.dark.negVSneut | -1.34226 | 0.000005 | -1.64318 | 0.000009 | -1.96359 | 0.000010 |
| large.bright.negVSneut | -1.04391 | 0.000005 | -1.32523 | 0.000006 | -1.63315 | 0.000006 |
| small.bright.negVSneut | -1.30534 | 0.000005 | -1.60379 | 0.000009 | -1.92265 | 0.000010 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in N1 amplitude:
- large size, high contrast: 0.173461;
- small size, high contrast: 0.193365;
- large size, low contrast: 0.265741;
- small size, low contrast: 0.201133.
Exploratory model selection
for (k in 1:length(scaling.factor)) {
N1.anovaBF <- anovaBF(
N1 ~ size * cont * emo + participant,
data.EEG.trial,
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(N1.anovaBF,
file = here::here(paste0("EEG/main/models/N1_anovaBF_", scaling.factor[k], ".rds")))
}compare.N1.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.N1.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
N1.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/N1_anovaBF_", scaling.factor[k], ".rds"))
)
compare.N1.anovaBF.temp[, k] <- N1.anovaBF@bayesFactor$bf
compare.N1.anovaBF.perc.err.temp[, k] <- N1.anovaBF@bayesFactor$error * 100
}
compare.N1.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.N1.anovaBF.temp[, 1], "nar.p.err" = compare.N1.anovaBF.perc.err.temp[, 1],
"med" = compare.N1.anovaBF.temp[, 2], "med.p.err" = compare.N1.anovaBF.perc.err.temp[, 2],
"wid" = compare.N1.anovaBF.temp[, 3], "wid.p.err" = compare.N1.anovaBF.perc.err.temp[, 3]
)
compare.N1.anovaBF <- compare.N1.anovaBF[order(compare.N1.anovaBF$med, decreasing = TRUE), ]
kable(compare.N1.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 2 | cont:emo | 3.73029 | 11.3004 | 4.21092 | 10.08546 | 4.41023 | 12.1467 |
| 3 | emo:size | 3.78233 | 10.9273 | 4.13584 | 9.92758 | 4.57307 | 22.7891 |
| 1 | cont:emo:size | 3.36395 | 13.3007 | 3.99916 | 9.90268 | 3.98102 | 11.7952 |
| 5 | emo | 3.34794 | 10.9888 | 3.81375 | 9.08760 | 4.15928 | 13.2988 |
| 7 | size | -41.23282 | 11.4377 | -40.69160 | 10.02464 | -40.54603 | 12.5904 |
| 4 | cont:size | -51.11622 | 12.8625 | -50.84828 | 10.26856 | -50.72175 | 12.5625 |
| 6 | cont | -127.35049 | 11.2607 | -126.85935 | 10.05205 | -126.72689 | 12.1201 |
Exploratory analyses further suggest that omitting contrast x emotion from the full model improves fitting by 67.418689 times. Removing size x emotion, size x contrast x emotion, and emotion also improves fitting by 62.541803, 54.552148, and 45.320135 times, respectively.
Conversely, omitting the factor size or the contrast x size interaction lowers the explanatory value of the resulting model by 4.70043e+17 and 1.21095e+22 times. Finally, omitting the factor contrast is maximally detrimental, as it would lower the explanatory value of the resulting model by 1.24255e+55 times.
EPN
summarySEwithin(
data.EEG.trial,
"EPN",
withinvars = c("size", "cont", "emo"),
idvar = "participant"
) %>%
kable(., digits = 2)| size | cont | emo | N | EPN | sd | se | ci |
|---|---|---|---|---|---|---|---|
| large | dark | negative | 2374 | 0.33 | 3.81 | 0.08 | 0.15 |
| large | dark | neutral | 2369 | 0.53 | 3.75 | 0.08 | 0.15 |
| large | bright | negative | 2373 | 0.27 | 3.93 | 0.08 | 0.16 |
| large | bright | neutral | 2364 | 0.58 | 3.88 | 0.08 | 0.16 |
| small | dark | negative | 2383 | 1.04 | 3.86 | 0.08 | 0.16 |
| small | dark | neutral | 2376 | 1.37 | 3.84 | 0.08 | 0.15 |
| small | bright | negative | 2365 | 1.15 | 4.01 | 0.08 | 0.16 |
| small | bright | neutral | 2363 | 1.09 | 4.11 | 0.08 | 0.17 |
compare.EPN.BF <- matrix(NA, 6, length(scaling.factor))
compare.EPN.perc.err <- matrix(NA, 6, length(scaling.factor))
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
EPN.sizeplusemo.BF <- lmBF(
EPN ~ size + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(EPN.sizeplusemo.BF,
file = here::here(paste0("EEG/main/models/EPN_sizeplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size and emotion
EPN.sizebyemo.BF <- lmBF(
EPN ~ size * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(EPN.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/EPN_sizebyemo_BF_", scaling.factor[k], ".rds")))
### main effects of contrast and emotion
EPN.contplusemo.BF <- lmBF(
EPN ~ cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(EPN.contplusemo.BF,
file = here::here(paste0("EEG/main/models/EPN_contplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of contrast and emotion
EPN.contbyemo.BF <- lmBF(
EPN ~ cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(EPN.contbyemo.BF,
file = here::here(paste0("EEG/main/models/EPN_contbyemo_BF_", scaling.factor[k], ".rds")))
### main effects of size, contrast, and emotion
EPN.sizepluscontplusemo.BF <- lmBF(
EPN ~ size + cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(EPN.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/EPN_sizepluscontplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size, contrast, and emotion
EPN.sizebycontbyemo.BF <- lmBF(
EPN ~ size * cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(EPN.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/EPN_sizebycontbyemo_BF_", scaling.factor[k], ".rds")))
}compare.EPN.BF <- matrix(NA, 6, length(scaling.factor))
compare.EPN.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) { # loop through scaling factors
### load models
EPN.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/EPN_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
EPN.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/EPN_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
EPN.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/EPN_contplusemo_BF_", scaling.factor[k], ".rds"))
)
EPN.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/EPN_contbyemo_BF_", scaling.factor[k], ".rds"))
)
EPN.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/EPN_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
EPN.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/EPN_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
# BFs
compare.EPN.BF[, k] <- c(
EPN.sizeplusemo.BF@bayesFactor$bf,
EPN.sizebyemo.BF@bayesFactor$bf,
EPN.contplusemo.BF@bayesFactor$bf,
EPN.contbyemo.BF@bayesFactor$bf,
EPN.sizepluscontplusemo.BF@bayesFactor$bf,
EPN.sizebycontbyemo.BF@bayesFactor$bf
)
# percentage of error
compare.EPN.perc.err[, k] <- c(
EPN.sizeplusemo.BF@bayesFactor$error * 100,
EPN.sizebyemo.BF@bayesFactor$error * 100,
EPN.contplusemo.BF@bayesFactor$error * 100,
EPN.contbyemo.BF@bayesFactor$error * 100,
EPN.sizepluscontplusemo.BF@bayesFactor$error * 100,
EPN.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
# summary confirmatory analysis
compare.EPN <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.EPN.BF[, 1], "nar.p.err" = compare.EPN.perc.err[, 1],
"med" = compare.EPN.BF[, 2], "med.p.err" = compare.EPN.perc.err[, 2],
"wid" = compare.EPN.BF[, 3], "wid.p.err" = compare.EPN.perc.err[, 3]
)
compare.EPN <- compare.EPN[order(compare.EPN$med, decreasing = TRUE), ]
kable(compare.EPN)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 1 | size + emo | 1326.49 | 2.31222 | 1325.89 | 2.40473 | 1325.18 | 2.33687 |
| 2 | size x emo | 1323.44 | 2.87939 | 1322.39 | 3.00713 | 1321.36 | 2.97128 |
| 5 | size + cont + emo | 1322.75 | 3.04108 | 1321.75 | 3.03328 | 1320.76 | 3.06843 |
| 6 | size x cont x emo | 1312.66 | 6.31408 | 1310.35 | 7.02359 | 1307.86 | 5.63571 |
| 3 | contr + emo | 1232.33 | 2.43123 | 1231.61 | 2.42752 | 1230.89 | 2.44795 |
| 4 | cont x emo | 1229.29 | 2.93835 | 1228.24 | 3.03208 | 1227.26 | 3.00893 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size + emo ought to be preferred.
The best model (size + emo) explains the observed data 33.058994 times better than the second best model (size x emo).
Paired comparisons
EPN.posthocBF <-
data.EEG.trial %>%
select(participant, cond, EPN) %>%
summarySEwithin(data = ., measurevar = "EPN", withinvars = c("participant", "cond")) %>%
split(.$cond)
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
EPN.posthocBF.large.dark.negVSneut <- ttestBF(
EPN.posthocBF$`negative large dark`$EPN,
EPN.posthocBF$`neutral large dark`$EPN,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(EPN.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/EPN_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds")))
### small, dark, negative vs. neutral
EPN.posthocBF.small.dark.negVSneut <- ttestBF(
EPN.posthocBF$`negative small dark`$EPN,
EPN.posthocBF$`neutral small dark`$EPN,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(EPN.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/EPN_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds")))
### large, bright, negative vs. neutral
EPN.posthocBF.large.bright.negVSneut <- ttestBF(
EPN.posthocBF$`negative large bright`$EPN,
EPN.posthocBF$`neutral large bright`$EPN,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(EPN.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/EPN_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds")))
### small, bright, negative vs. neutral
EPN.posthocBF.small.bright.negVSneut <- ttestBF(
EPN.posthocBF$`negative small bright`$EPN,
EPN.posthocBF$`neutral small bright`$EPN,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(EPN.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/EPN_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds")))
}compare.EPN.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.EPN.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
EPN.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/EPN_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
EPN.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/EPN_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
EPN.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/EPN_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
EPN.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/EPN_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
compare.EPN.posthocBF[, k] <- c(
EPN.posthocBF.large.dark.negVSneut@bayesFactor$bf,
EPN.posthocBF.small.dark.negVSneut@bayesFactor$bf,
EPN.posthocBF.large.bright.negVSneut@bayesFactor$bf,
EPN.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
compare.EPN.posthocBF.perc.err[, k] <- c(
EPN.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
EPN.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
EPN.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
EPN.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
compare.EPN.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.EPN.posthocBF[, 1], "nar.p.err" = compare.EPN.posthocBF.perc.err[, 1],
"med" = compare.EPN.posthocBF[, 2], "med.p.err" = compare.EPN.posthocBF.perc.err[, 2],
"wid" = compare.EPN.posthocBF[, 3], "wid.p.err" = compare.EPN.posthocBF.perc.err[, 3]
)
kable(compare.EPN.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -0.400217 | 0.000004 | -0.641569 | 0.000003 | -0.923346 | 0.000002 |
| small.dark.negVSneut | 2.410784 | 0.000000 | 2.312172 | 0.000808 | 2.132030 | 0.000000 |
| large.bright.negVSneut | 1.811765 | 0.000114 | 1.686604 | 0.000000 | 1.486398 | 0.000000 |
| small.bright.negVSneut | -1.369311 | 0.000005 | -1.672038 | 0.000009 | -1.993603 | 0.000011 |
Follow-up contrasts reveal that emotion may explain changes in EPN amplitude when words are presented in small font and high contrast (10.096328), as well as large font and low contrast (5.40111).
Exploratory model selection
for (k in 1:length(scaling.factor)) {
EPN.anovaBF <- anovaBF(
EPN ~ size * cont * emo + participant,
data.EEG.trial,
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(EPN.anovaBF,
file = here::here(paste0("EEG/main/models/EPN_anovaBF_", scaling.factor[k], ".rds")))
}compare.EPN.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.EPN.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
EPN.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/EPN_anovaBF_", scaling.factor[k], ".rds"))
)
compare.EPN.anovaBF.temp[, k] <- EPN.anovaBF@bayesFactor$bf
compare.EPN.anovaBF.perc.err.temp[, k] <- EPN.anovaBF@bayesFactor$error * 100
}
compare.EPN.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.EPN.anovaBF.temp[, 1], "nar.p.err" = compare.EPN.anovaBF.perc.err.temp[, 1],
"med" = compare.EPN.anovaBF.temp[, 2], "med.p.err" = compare.EPN.anovaBF.perc.err.temp[, 2],
"wid" = compare.EPN.anovaBF.temp[, 3], "wid.p.err" = compare.EPN.anovaBF.perc.err.temp[, 3]
)
compare.EPN.anovaBF <- compare.EPN.anovaBF[order(compare.EPN.anovaBF$med, decreasing = TRUE), ]
kable(compare.EPN.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 6 | cont | 3.76437 | 11.2963 | 3.95438 | 10.55308 | 4.55411 | 8.35918 |
| 4 | cont:size | 3.48978 | 11.5879 | 3.78341 | 9.74697 | 4.33640 | 8.71177 |
| 3 | emo:size | 3.13453 | 10.5641 | 3.42454 | 9.60220 | 3.90016 | 8.24739 |
| 2 | cont:emo | 2.91558 | 11.6178 | 3.13734 | 9.36192 | 3.73162 | 9.60535 |
| 1 | cont:emo:size | 0.57099 | 11.4056 | 0.80486 | 9.54323 | 1.18533 | 7.98465 |
| 5 | emo | -2.91527 | 10.6151 | -2.53273 | 9.72118 | -2.03308 | 9.34033 |
| 7 | size | -90.39216 | 11.6568 | -90.41695 | 9.45235 | -89.82763 | 8.50455 |
Exploratory analyses further suggest that omitting the factor contrast from the full model improves fitting by 52.163543 times. Removing contrast x size, size x emotion, contrast x emotion, and size x contrast x emotion also improves fitting by 43.965611, 30.708517, 23.042374, and 2.236383 times, respectively.
Conversely, omitting the factor emotion lowers the explanatory value of the resulting model by 12.587773 times. Finally, omitting the factor size is maximally detrimental, as it would lower the explanatory value of the resulting model by 1.85175e+39 times.
LPP
summarySEwithin(
data.EEG.trial,
"LPP",
withinvars = c("size", "cont", "emo"),
idvar = "participant"
) %>%
kable(., digits = 2)| size | cont | emo | N | LPP | sd | se | ci |
|---|---|---|---|---|---|---|---|
| large | dark | negative | 2374 | 0.61 | 3.04 | 0.06 | 0.12 |
| large | dark | neutral | 2369 | 0.60 | 2.95 | 0.06 | 0.12 |
| large | bright | negative | 2373 | 0.71 | 3.03 | 0.06 | 0.12 |
| large | bright | neutral | 2364 | 0.81 | 3.04 | 0.06 | 0.12 |
| small | dark | negative | 2383 | 0.76 | 3.09 | 0.06 | 0.12 |
| small | dark | neutral | 2376 | 0.79 | 3.09 | 0.06 | 0.12 |
| small | bright | negative | 2365 | 1.35 | 3.31 | 0.07 | 0.13 |
| small | bright | neutral | 2363 | 1.17 | 3.36 | 0.07 | 0.14 |
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
LPP.sizeplusemo.BF <- lmBF(
LPP ~ size + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(LPP.sizeplusemo.BF,
file = here::here(paste0("EEG/main/models/LPP_sizeplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size and emotion
LPP.sizebyemo.BF <- lmBF(
LPP ~ size * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(LPP.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/LPP_sizebyemo_BF_", scaling.factor[k], ".rds")))
### main effects of contrast and emotion
LPP.contplusemo.BF <- lmBF(
LPP ~ cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(LPP.contplusemo.BF,
file = here::here(paste0("EEG/main/models/LPP_contplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of contrast and emotion
LPP.contbyemo.BF <- lmBF(
LPP ~ cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(LPP.contbyemo.BF,
file = here::here(paste0("EEG/main/models/LPP_contbyemo_BF_", scaling.factor[k], ".rds")))
### main effects of size, contrast, and emotion
LPP.sizepluscontplusemo.BF <- lmBF(
LPP ~ size + cont + emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(LPP.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/LPP_sizepluscontplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size, contrast, and emotion
LPP.sizebycontbyemo.BF <- lmBF(
LPP ~ size * cont * emo + participant,
data.EEG.trial,
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(LPP.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/LPP_sizebycontbyemo_BF_", scaling.factor[k], ".rds")))
}compare.LPP.BF <- matrix(NA, 6, length(scaling.factor))
compare.LPP.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) { # loop through scaling factors
### load models
LPP.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/LPP_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
LPP.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/LPP_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
LPP.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/LPP_contplusemo_BF_", scaling.factor[k], ".rds"))
)
LPP.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/LPP_contbyemo_BF_", scaling.factor[k], ".rds"))
)
LPP.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/LPP_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
LPP.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/LPP_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
# BFs
compare.LPP.BF[, k] <- c(
LPP.sizeplusemo.BF@bayesFactor$bf,
LPP.sizebyemo.BF@bayesFactor$bf,
LPP.contplusemo.BF@bayesFactor$bf,
LPP.contbyemo.BF@bayesFactor$bf,
LPP.sizepluscontplusemo.BF@bayesFactor$bf,
LPP.sizebycontbyemo.BF@bayesFactor$bf
)
# percentage of error
compare.LPP.perc.err[, k] <- c(
LPP.sizeplusemo.BF@bayesFactor$error * 100,
LPP.sizebyemo.BF@bayesFactor$error * 100,
LPP.contplusemo.BF@bayesFactor$error * 100,
LPP.contbyemo.BF@bayesFactor$error * 100,
LPP.sizepluscontplusemo.BF@bayesFactor$error * 100,
LPP.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
# summary confirmatory analysis
compare.LPP <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.LPP.BF[, 1], "nar.p.err" = compare.LPP.perc.err[, 1],
"med" = compare.LPP.BF[, 2], "med.p.err" = compare.LPP.perc.err[, 2],
"wid" = compare.LPP.BF[, 3], "wid.p.err" = compare.LPP.perc.err[, 3]
)
compare.LPP <- compare.LPP[order(compare.LPP$med, decreasing = TRUE), ]
kable(compare.LPP)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | size + cont + emo | 520.406 | 11.22651 | 519.446 | 17.5976 | 518.560 | 11.5647 |
| 6 | size x cont x emo | 516.093 | 17.08671 | 513.643 | 15.5526 | 511.146 | 21.9664 |
| 1 | size + emo | 496.161 | 9.78495 | 495.544 | 9.3165 | 494.769 | 11.5264 |
| 3 | contr + emo | 494.110 | 11.09732 | 493.459 | 10.7985 | 492.771 | 12.4911 |
| 2 | size x emo | 493.421 | 11.97174 | 492.545 | 16.8120 | 491.177 | 12.1359 |
| 4 | cont x emo | 490.577 | 12.05388 | 489.647 | 11.9476 | 488.485 | 11.5656 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size + cont + emo ought to be preferred.
The best model (size + cont + emo) explains the observed data 331.122922 times better than the second best model (size x cont x emo).
Paired comparisons
LPP.posthocBF <-
data.EEG.trial %>%
select(participant, cond, LPP) %>%
summarySEwithin(data = ., measurevar = "LPP", withinvars = c("participant", "cond")) %>%
split(.$cond)
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
LPP.posthocBF.large.dark.negVSneut <- ttestBF(
LPP.posthocBF$`negative large dark`$LPP,
LPP.posthocBF$`neutral large dark`$LPP,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(LPP.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/LPP_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds")))
### small, dark, negative vs. neutral
LPP.posthocBF.small.dark.negVSneut <- ttestBF(
LPP.posthocBF$`negative small dark`$LPP,
LPP.posthocBF$`neutral small dark`$LPP,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(LPP.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/LPP_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds")))
### large, bright, negative vs. neutral
LPP.posthocBF.large.bright.negVSneut <- ttestBF(
LPP.posthocBF$`negative large bright`$LPP,
LPP.posthocBF$`neutral large bright`$LPP,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(LPP.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/LPP_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds")))
### small, bright, negative vs. neutral
LPP.posthocBF.small.bright.negVSneut <- ttestBF(
LPP.posthocBF$`negative small bright`$LPP,
LPP.posthocBF$`neutral small bright`$LPP,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(LPP.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/LPP_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds")))
}compare.LPP.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.LPP.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
LPP.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/LPP_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
LPP.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/LPP_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
LPP.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/LPP_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
LPP.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/LPP_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
compare.LPP.posthocBF[, k] <- c(
LPP.posthocBF.large.dark.negVSneut@bayesFactor$bf,
LPP.posthocBF.small.dark.negVSneut@bayesFactor$bf,
LPP.posthocBF.large.bright.negVSneut@bayesFactor$bf,
LPP.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
compare.LPP.posthocBF.perc.err[, k] <- c(
LPP.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
LPP.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
LPP.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
LPP.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
compare.LPP.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.LPP.posthocBF[, 1], "nar.p.err" = compare.LPP.posthocBF.perc.err[, 1],
"med" = compare.LPP.posthocBF[, 2], "med.p.err" = compare.LPP.posthocBF.perc.err[, 2],
"wid" = compare.LPP.posthocBF[, 3], "wid.p.err" = compare.LPP.posthocBF.perc.err[, 3]
)
kable(compare.LPP.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.458985 | 0.000005 | -1.767756 | 0.000010 | -2.093136 | 0.000012 |
| small.dark.negVSneut | -1.388719 | 0.000005 | -1.692748 | 0.000009 | -2.015137 | 0.000011 |
| large.bright.negVSneut | -0.908117 | 0.000005 | -1.180755 | 0.000005 | -1.483057 | 0.000005 |
| small.bright.negVSneut | -0.007331 | 0.000003 | -0.225745 | 0.000001 | -0.492124 | 0.000001 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in LPP amplitude:
- large size, high contrast: 0.170716;
- small size, high contrast: 0.184013;
- large size, low contrast: 0.307047;
- small size, low contrast: 0.797921.
Exploratory model selection
for (k in 1:length(scaling.factor)) {
LPP.anovaBF <- anovaBF(
LPP ~ size * cont * emo + participant,
data.EEG.trial,
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(LPP.anovaBF,
file = here::here(paste0("EEG/main/models/LPP_anovaBF_", scaling.factor[k], ".rds")))
}compare.LPP.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.LPP.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
LPP.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/LPP_anovaBF_", scaling.factor[k], ".rds"))
)
compare.LPP.anovaBF.temp[, k] <- LPP.anovaBF@bayesFactor$bf
compare.LPP.anovaBF.perc.err.temp[, k] <- LPP.anovaBF@bayesFactor$error * 100
}
compare.LPP.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.LPP.anovaBF.temp[, 1], "nar.p.err" = compare.LPP.anovaBF.perc.err.temp[, 1],
"med" = compare.LPP.anovaBF.temp[, 2], "med.p.err" = compare.LPP.anovaBF.perc.err.temp[, 2],
"wid" = compare.LPP.anovaBF.temp[, 3], "wid.p.err" = compare.LPP.anovaBF.perc.err.temp[, 3]
)
compare.LPP.anovaBF <- compare.LPP.anovaBF[order(compare.LPP.anovaBF$med, decreasing = TRUE), ]
kable(compare.LPP.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | emo | 3.97403 | 28.5906 | 4.38369 | 26.1832 | 5.11209 | 27.4134 |
| 2 | cont:emo | 3.50601 | 29.2593 | 3.75328 | 28.9674 | 4.17262 | 25.7595 |
| 3 | emo:size | 2.69398 | 30.2241 | 3.05430 | 27.2681 | 3.61779 | 28.1896 |
| 1 | cont:emo:size | 1.61271 | 30.2210 | 1.75300 | 25.6822 | 2.70414 | 30.1085 |
| 4 | cont:size | -3.91812 | 28.1330 | -3.50266 | 27.6769 | -2.90837 | 26.5374 |
| 6 | cont | -24.71072 | 25.8800 | -24.27263 | 29.5110 | -22.94021 | 29.9496 |
| 7 | size | -26.34967 | 31.3473 | -25.54122 | 41.2221 | -25.61069 | 28.5671 |
Exploratory analyses further suggest that omitting the factor emotion from the full model improves fitting by 80.132853 times. Removing contrast x emotion, size x emotion, and size x contrast x emotion also improves fitting by 42.660828, 21.206335, 5.771892 times, respectively.
Conversely, omitting the contrast x size interaction or the factor contrast lowers the explanatory value of the resulting model by 33.203549 and 3.47912e+10 times, respectively. Finally, omitting the factor size is maximally detrimental, as it would lower the explanatory value of the resulting model by 1.23712e+11 times.
SUPPLEMENTARY ANALYSIS
Visual inspection of the waveforms revealed unexpected latency shifts at the level of P1 and N1 components. This could be a potential source of noise when analyzing mean amplitude values. Therefore, we ran the same analyses as above but using, as dependent variables, peak amplitude and latency values of P1 and N1.
# load data
data.P1.peakamp <- read.table(here::here("EEG/main/data/P1_locpeakamp.txt"), header = TRUE)
data.P1.peaklat <- read.table(here::here("EEG/main/data/P1_locpeaklat.txt"), header = TRUE)
data.N1.peakamp <- read.table(here::here("EEG/main/data/N1_locpeakamp.txt"), header = TRUE)
data.N1.peaklat <- read.table(here::here("EEG/main/data/N1_locpeaklat.txt"), header = TRUE)
data.peaks <-
rbind(data.P1.peakamp, data.P1.peaklat, data.N1.peakamp, data.N1.peaklat) %>%
mutate(
var = rep(c("peak.amp", "peak.lat", "peak.amp", "peak.lat"), each = 320), # peak amplitude or latency?
bini = factor(bini), # convert condition numbers as factors
size = revalue(
bini, # main effect of size
c(
"1" = "large", "2" = "small", "3" = "large", "4" = "small",
"5" = "large", "6" = "small", "7" = "large", "8" = "small"
)
),
size = relevel(size, ref = "large"), # re-reference size to large
cont = revalue(
bini, # main effect of contrast
c(
"1" = "dark", "2" = "dark", "3" = "bright", "4" = "bright",
"5" = "dark", "6" = "dark", "7" = "bright", "8" = "bright"
)
),
cont = relevel(cont, ref = "dark"), # re-reference contrast to dark
emo = revalue(
bini, # main effect of emotion
c(
"1" = "negative", "2" = "negative", "3" = "negative", "4" = "negative",
"5" = "neutral", "6" = "neutral", "7" = "neutral", "8" = "neutral"
)
),
emo = relevel(emo, ref = "neutral") # re-reference emotion to neutral
) %>%
unite(visualfeats, c("size", "cont"), sep = " ", remove = FALSE) %>% # for graph
mutate(visualfeats = as.factor(ifelse(visualfeats == "NA NA", "localizer", visualfeats))) %>%
# select, reorder, and rename variables
dplyr::select(participant = ERPset, component = chlabel, var, condition = binlabel, size, cont, emo, visualfeats, value)
rm(data.P1.peakamp, data.P1.peaklat, data.N1.peakamp, data.N1.peaklat) # remove unnecessary data framesP1
Peak amplitude
Peak amplitude: positive value larger than the 3 points (~10 ms at 256 Hz sampling rate) on either side of the peak.
Time window: 66-148 ms post-stimulus onset.
Electrode cluster: P7 P9 PO7 O1 O2 PO8 P8 P10.
# summarize data
summary.P1.peakamp <-
summarySEwithin(
subset(data.peaks, component == "P1" & var == "peak.amp"),
"value",
withinvars = c("size", "cont", "emo", "visualfeats"),
idvar = "participant"
)
kable(summary.P1.peakamp)| size | cont | emo | visualfeats | N | value | sd | se | ci |
|---|---|---|---|---|---|---|---|---|
| large | dark | neutral | large dark | 40 | 3.35137 | 1.025301 | 0.162114 | 0.327907 |
| large | dark | negative | large dark | 40 | 3.29277 | 1.064229 | 0.168269 | 0.340357 |
| large | bright | neutral | large bright | 40 | 3.00358 | 0.940474 | 0.148702 | 0.300778 |
| large | bright | negative | large bright | 40 | 3.13585 | 0.873157 | 0.138058 | 0.279249 |
| small | dark | neutral | small dark | 40 | 2.66942 | 0.941481 | 0.148861 | 0.301100 |
| small | dark | negative | small dark | 40 | 2.90865 | 0.833393 | 0.131771 | 0.266532 |
| small | bright | neutral | small bright | 40 | 1.29638 | 1.664701 | 0.263212 | 0.532397 |
| small | bright | negative | small bright | 40 | 1.46732 | 1.222649 | 0.193318 | 0.391022 |
# pirateplot
pirateplot(
formula = value ~ emo + visualfeats,
data = subset(data.peaks, component == "P1" & var == "peak.amp"),
main = "P1 peak amplitude",
xlim = NULL,
xlab = "",
ylim = c(-1, 11),
ylab = expression(paste("amplitude (", mu, "V)")),
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
P1.peakamp.sizeplusemo.BF <- lmBF(
value ~ size + emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.sizeplusemo.BF, # save model
file = here::here(paste0("EEG/main/models/P1_peakamp_sizeplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size and emotion
P1.peakamp.sizebyemo.BF <- lmBF(
value ~ size * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/P1_peakamp_sizebyemo_BF_", scaling.factor[k], ".rds")))
### main effects of contrast and emotion
P1.peakamp.contplusemo.BF <- lmBF(
value ~ cont + emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.contplusemo.BF,
file = here::here(paste0("EEG/main/models/P1_peakamp_contplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of contrast and emotion
P1.peakamp.contbyemo.BF <- lmBF(
value ~ cont * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.contbyemo.BF,
file = here::here(paste0("EEG/main/models/P1_peakamp_contbyemo_BF_", scaling.factor[k], ".rds")))
### main effects of size, contrast, and emotion
P1.peakamp.sizepluscontplusemo.BF <- lmBF(
value ~ size + cont + emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/P1_peakamp_sizepluscontplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size, contrast, and emotion
P1.peakamp.sizebycontbyemo.BF <- lmBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/P1_peakamp_sizebycontbyemo_BF_", scaling.factor[k], ".rds")))
}compare.P1.peakamp.BF <- matrix(NA, 6, length(scaling.factor))
compare.P1.peakamp.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
P1.peakamp.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.peakamp.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
P1.peakamp.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_contplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.peakamp.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_contbyemo_BF_", scaling.factor[k], ".rds"))
)
P1.peakamp.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.peakamp.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
compare.P1.peakamp.BF[, k] <- c(
P1.peakamp.sizeplusemo.BF@bayesFactor$bf,
P1.peakamp.sizebyemo.BF@bayesFactor$bf,
P1.peakamp.contplusemo.BF@bayesFactor$bf,
P1.peakamp.contbyemo.BF@bayesFactor$bf,
P1.peakamp.sizepluscontplusemo.BF@bayesFactor$bf,
P1.peakamp.sizebycontbyemo.BF@bayesFactor$bf
)
compare.P1.peakamp.perc.err[, k] <- c(
P1.peakamp.sizeplusemo.BF@bayesFactor$error * 100,
P1.peakamp.sizebyemo.BF@bayesFactor$error * 100,
P1.peakamp.contplusemo.BF@bayesFactor$error * 100,
P1.peakamp.contbyemo.BF@bayesFactor$error * 100,
P1.peakamp.sizepluscontplusemo.BF@bayesFactor$error * 100,
P1.peakamp.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
compare.P1.peakamp <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.P1.peakamp.BF[, 1], "nar.p.err" = compare.P1.peakamp.perc.err[, 1],
"med" = compare.P1.peakamp.BF[, 2], "med.p.err" = compare.P1.peakamp.perc.err[, 2],
"wid" = compare.P1.peakamp.BF[, 3], "wid.p.err" = compare.P1.peakamp.perc.err[, 3]
)
compare.P1.peakamp <- compare.P1.peakamp[order(compare.P1.peakamp$med, decreasing = TRUE), ]
kable(compare.P1.peakamp)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 6 | size x cont x emo | 113.0472 | 2.082444 | 111.6635 | 2.226273 | 109.9492 | 2.505049 |
| 5 | size + cont + emo | 110.3954 | 0.972598 | 109.9963 | 0.996991 | 109.4234 | 1.068096 |
| 1 | size + emo | 94.8978 | 0.864897 | 94.5627 | 0.880644 | 94.1140 | 0.904550 |
| 2 | size x emo | 93.2742 | 1.068691 | 92.6474 | 1.096823 | 91.8743 | 1.137737 |
| 3 | contr + emo | 82.2755 | 0.853039 | 81.8464 | 0.885007 | 81.3105 | 0.927919 |
| 4 | cont x emo | 80.5544 | 1.060617 | 79.7795 | 1.120119 | 78.9037 | 1.169013 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size x cont x emo ought to be preferred.
The best model (size x cont x emo) explains the observed data 5.29724 times better than the second best model (size + cont + emo).
Paired comparisons
P1.peakamp.posthocBF <-
subset(data.peaks, component == "P1" & var == "peak.amp") %>%
select(participant, condition, value) %>%
split(.$condition)
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
P1.peakamp.posthocBF.large.dark.negVSneut <- ttestBF(
P1.peakamp.posthocBF$`negativeLargeDark`$value,
P1.peakamp.posthocBF$`neutralLargeDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peakamp.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds")))
### small, dark, negative vs. neutral
P1.peakamp.posthocBF.small.dark.negVSneut <- ttestBF(
P1.peakamp.posthocBF$`negativeSmallDark`$value,
P1.peakamp.posthocBF$`neutralSmallDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peakamp.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds")))
### large, bright, negative vs. neutral
P1.peakamp.posthocBF.large.bright.negVSneut <- ttestBF(
P1.peakamp.posthocBF$`negativeLargeBright`$value,
P1.peakamp.posthocBF$`neutralLargeBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peakamp.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds")))
### small, bright, negative vs. neutral
P1.peakamp.posthocBF.small.bright.negVSneut <- ttestBF(
P1.peakamp.posthocBF$`negativeSmallBright`$value,
P1.peakamp.posthocBF$`neutralSmallBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peakamp.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds")))
}compare.P1.peakamp.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.P1.peakamp.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
P1.peakamp.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
P1.peakamp.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
P1.peakamp.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
P1.peakamp.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### model comparison
compare.P1.peakamp.posthocBF[, k] <- c(
P1.peakamp.posthocBF.large.dark.negVSneut@bayesFactor$bf,
P1.peakamp.posthocBF.small.dark.negVSneut@bayesFactor$bf,
P1.peakamp.posthocBF.large.bright.negVSneut@bayesFactor$bf,
P1.peakamp.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
# percentage of error
compare.P1.peakamp.posthocBF.perc.err[, k] <- c(
P1.peakamp.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
P1.peakamp.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
P1.peakamp.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
P1.peakamp.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
# summary
compare.P1.peakamp.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.P1.peakamp.posthocBF[, 1], "nar.p.err" = compare.P1.peakamp.posthocBF.perc.err[, 1],
"med" = compare.P1.peakamp.posthocBF[, 2], "med.p.err" = compare.P1.peakamp.posthocBF.perc.err[, 2],
"wid" = compare.P1.peakamp.posthocBF[, 3], "wid.p.err" = compare.P1.peakamp.posthocBF.perc.err[, 3]
)
kable(compare.P1.peakamp.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.32816 | 0.000005 | -1.628132 | 0.000009 | -1.947953 | 0.000010 |
| small.dark.negVSneut | -0.07760 | 0.000004 | -0.300039 | 0.000002 | -0.569142 | 0.000001 |
| large.bright.negVSneut | -1.22201 | 0.000005 | -1.514945 | 0.000008 | -1.830291 | 0.000008 |
| small.bright.negVSneut | -1.19346 | 0.000005 | -1.484519 | 0.000008 | -1.798667 | 0.000008 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in P1 peak amplitude:
- large size, high contrast: 0.196296;
- small size, high contrast: 0.74079;
- large size, low contrast: 0.21982;
- small size, low contrast: 0.226611.
Exploratory model selection
for (k in 1:length(scaling.factor)) {
P1.peakamp.anovaBF <- anovaBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.amp"),
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(P1.peakamp.anovaBF,
file = here::here(paste0("EEG/main/models/P1_peakamp_anovaBF_", scaling.factor[k], ".rds")))
}compare.P1.peakamp.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.P1.peakamp.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load model
P1.peakamp.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/P1_peakamp_anovaBF_", scaling.factor[k], ".rds"))
)
compare.P1.peakamp.anovaBF.temp[, k] <- P1.peakamp.anovaBF@bayesFactor$bf
compare.P1.peakamp.anovaBF.perc.err.temp[, k] <- P1.peakamp.anovaBF@bayesFactor$error * 100
}
compare.P1.peakamp.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.P1.peakamp.anovaBF.temp[, 1], "nar.p.err" = compare.P1.peakamp.anovaBF.perc.err.temp[, 1],
"med" = compare.P1.peakamp.anovaBF.temp[, 2], "med.p.err" = compare.P1.peakamp.anovaBF.perc.err.temp[, 2],
"wid" = compare.P1.peakamp.anovaBF.temp[, 3], "wid.p.err" = compare.P1.peakamp.anovaBF.perc.err.temp[, 3]
)
compare.P1.peakamp.anovaBF <- compare.P1.peakamp.anovaBF[order(compare.P1.peakamp.anovaBF$med, decreasing = TRUE), ]
kable(compare.P1.peakamp.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 2 | cont:emo | 1.72525 | 2.53469 | 2.01233 | 2.84692 | 2.39872 | 3.21703 |
| 5 | emo | 1.65298 | 2.53784 | 1.92999 | 2.88048 | 2.34748 | 3.24195 |
| 3 | emo:size | 1.58722 | 2.52966 | 1.81079 | 2.83664 | 2.20254 | 3.25555 |
| 1 | cont:emo:size | 1.35922 | 2.54836 | 1.60462 | 2.87772 | 1.96946 | 3.29696 |
| 4 | cont:size | -7.33879 | 2.60862 | -7.36829 | 2.93751 | -7.16748 | 3.24773 |
| 6 | cont | -16.67293 | 2.65234 | -16.72574 | 2.93242 | -16.59378 | 3.23379 |
| 7 | size | -30.03498 | 2.65536 | -30.21739 | 3.00605 | -30.15072 | 3.38574 |
Exploratory analyses suggest that omitting contrast x emotion from the full model improves fitting by 7.480755 times. Removing emotion, size x emotion, and size x contrast x emotion also improves fitting by 6.889455, 6.115285, and 4.975987 times, respectively.
Conversely, omitting contrast x size or contrast lowers the explanatory value of the resulting model by 1584.929377 and 1.8361e+07 times, respectively. Finally, omitting the factor size is maximally detrimental, as it would lower the explanatory value of the resulting model by 1.32815e+13 times.
Peak latency
Peak latency: time (in ms) of the positive value larger than the 3 points (~10 ms at 256 Hz sampling rate) on either side of the peak.
Time window: 66-148 ms post-stimulus onset.
Electrode cluster: P7 P9 PO7 O1 O2 PO8 P8 P10.
summary.P1.peaklat <-
summarySEwithin(subset(data.peaks, component == "P1" & var == "peak.lat"),
"value",
withinvars = c("size", "cont", "emo", "visualfeats"),
idvar = "participant"
)
kable(summary.P1.peaklat)| size | cont | emo | visualfeats | N | value | sd | se | ci |
|---|---|---|---|---|---|---|---|---|
| large | dark | neutral | large dark | 40 | 103.711 | 13.59364 | 2.14934 | 4.34746 |
| large | dark | negative | large dark | 40 | 102.344 | 15.28566 | 2.41687 | 4.88859 |
| large | bright | neutral | large bright | 40 | 123.633 | 21.13074 | 3.34106 | 6.75794 |
| large | bright | negative | large bright | 40 | 126.270 | 17.38954 | 2.74953 | 5.56145 |
| small | dark | neutral | small dark | 40 | 113.184 | 9.92574 | 1.56940 | 3.17441 |
| small | dark | negative | small dark | 40 | 116.699 | 13.39870 | 2.11852 | 4.28511 |
| small | bright | neutral | small bright | 40 | 115.430 | 27.00762 | 4.27028 | 8.63745 |
| small | bright | negative | small bright | 40 | 116.992 | 22.47403 | 3.55346 | 7.18754 |
# pirateplot
pirateplot(
formula = value ~ emo + visualfeats,
data = subset(data.peaks, component == "P1" & var == "peak.lat"),
main = "P1 peak latency",
xlim = NULL,
xlab = "",
ylim = c(60, 150),
ylab = "time (ms)",
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
P1.peaklat.sizeplusemo.BF <- lmBF(
value ~ size + emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.sizeplusemo.BF, # save model
file = here::here(paste0("EEG/main/models/P1_peaklat_sizeplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size and emotion
P1.peaklat.sizebyemo.BF <- lmBF(
value ~ size * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/P1_peaklat_sizebyemo_BF_", scaling.factor[k], ".rds")))
### main effects of contrast and emotion
P1.peaklat.contplusemo.BF <- lmBF(
value ~ cont + emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.contplusemo.BF,
file = here::here(paste0("EEG/main/models/P1_peaklat_contplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of contrast and emotion
P1.peaklat.contbyemo.BF <- lmBF(
value ~ cont * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.contbyemo.BF,
file = here::here(paste0("EEG/main/models/P1_peaklat_contbyemo_BF_", scaling.factor[k], ".rds")))
### main effects of size, contrast, and emotion
P1.peaklat.sizepluscontplusemo.BF <- lmBF(
value ~ size + cont + emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/P1_peaklat_sizepluscontplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size, contrast, and emotion
P1.peaklat.sizebycontbyemo.BF <- lmBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/P1_peaklat_sizebycontbyemo_BF_", scaling.factor[k], ".rds")))
}compare.P1.peaklat.BF <- matrix(NA, 6, length(scaling.factor))
compare.P1.peaklat.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
P1.peaklat.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.peaklat.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
P1.peaklat.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_contplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.peaklat.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_contbyemo_BF_", scaling.factor[k], ".rds"))
)
P1.peaklat.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
P1.peaklat.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
compare.P1.peaklat.BF[, k] <- c(
P1.peaklat.sizeplusemo.BF@bayesFactor$bf,
P1.peaklat.sizebyemo.BF@bayesFactor$bf,
P1.peaklat.contplusemo.BF@bayesFactor$bf,
P1.peaklat.contbyemo.BF@bayesFactor$bf,
P1.peaklat.sizepluscontplusemo.BF@bayesFactor$bf,
P1.peaklat.sizebycontbyemo.BF@bayesFactor$bf
)
compare.P1.peaklat.perc.err[, k] <- c(
P1.peaklat.sizeplusemo.BF@bayesFactor$error * 100,
P1.peaklat.sizebyemo.BF@bayesFactor$error * 100,
P1.peaklat.contplusemo.BF@bayesFactor$error * 100,
P1.peaklat.contbyemo.BF@bayesFactor$error * 100,
P1.peaklat.sizepluscontplusemo.BF@bayesFactor$error * 100,
P1.peaklat.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
compare.P1.peaklat <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.P1.peaklat.BF[, 1], "nar.p.err" = compare.P1.peaklat.perc.err[, 1],
"med" = compare.P1.peaklat.BF[, 2], "med.p.err" = compare.P1.peaklat.perc.err[, 2],
"wid" = compare.P1.peaklat.BF[, 3], "wid.p.err" = compare.P1.peaklat.perc.err[, 3]
)
compare.P1.peaklat <- compare.P1.peaklat[order(compare.P1.peaklat$med, decreasing = TRUE), ]
kable(compare.P1.peaklat)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 6 | size x cont x emo | 10.39443 | 3.93015 | 8.64674 | 4.27277 | 6.61916 | 4.98297 |
| 3 | contr + emo | 7.84221 | 1.90041 | 7.38171 | 1.95927 | 6.84259 | 2.03116 |
| 4 | cont x emo | 6.09696 | 2.25124 | 5.32543 | 2.44768 | 4.40629 | 2.55865 |
| 5 | size + cont + emo | 6.00173 | 2.28447 | 5.21400 | 2.38414 | 4.36215 | 2.59607 |
| 1 | size + emo | -4.87032 | 2.48111 | -5.49833 | 2.45132 | -6.18859 | 2.48785 |
| 2 | size x emo | -6.61149 | 2.99039 | -7.54807 | 3.01909 | -8.55482 | 3.00385 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size x cont x emo ought to be preferred.
The best model (size x cont x emo) explains the observed data 3.543206 times better than the second best model (contr + emo).
Paired comparisons
P1.peaklat.posthocBF <-
subset(data.peaks, component == "P1" & var == "peak.lat") %>%
select(participant, condition, value) %>%
split(.$condition)
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
P1.peaklat.posthocBF.large.dark.negVSneut <- ttestBF(
P1.peaklat.posthocBF$`negativeLargeDark`$value,
P1.peaklat.posthocBF$`neutralLargeDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peaklat.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds")))
### small, dark, negative vs. neutral
P1.peaklat.posthocBF.small.dark.negVSneut <- ttestBF(
P1.peaklat.posthocBF$`negativeSmallDark`$value,
P1.peaklat.posthocBF$`neutralSmallDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peaklat.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds")))
### large, bright, negative vs. neutral
P1.peaklat.posthocBF.large.bright.negVSneut <- ttestBF(
P1.peaklat.posthocBF$`negativeLargeBright`$value,
P1.peaklat.posthocBF$`neutralLargeBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peaklat.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds")))
### small, bright, negative vs. neutral
P1.peaklat.posthocBF.small.bright.negVSneut <- ttestBF(
P1.peaklat.posthocBF$`negativeSmallBright`$value,
P1.peaklat.posthocBF$`neutralSmallBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(P1.peaklat.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds")))
}compare.P1.peaklat.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.P1.peaklat.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
P1.peaklat.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
P1.peaklat.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
P1.peaklat.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
P1.peaklat.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### model comparison
compare.P1.peaklat.posthocBF[, k] <- c(
P1.peaklat.posthocBF.large.dark.negVSneut@bayesFactor$bf,
P1.peaklat.posthocBF.small.dark.negVSneut@bayesFactor$bf,
P1.peaklat.posthocBF.large.bright.negVSneut@bayesFactor$bf,
P1.peaklat.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
# percentage of error
compare.P1.peaklat.posthocBF.perc.err[, k] <- c(
P1.peaklat.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
P1.peaklat.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
P1.peaklat.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
P1.peaklat.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
# summary
compare.P1.peaklat.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.P1.peaklat.posthocBF[, 1], "nar.p.err" = compare.P1.peaklat.posthocBF.perc.err[, 1],
"med" = compare.P1.peaklat.posthocBF[, 2], "med.p.err" = compare.P1.peaklat.posthocBF.perc.err[, 2],
"wid" = compare.P1.peaklat.posthocBF[, 3], "wid.p.err" = compare.P1.peaklat.posthocBF.perc.err[, 3]
)
kable(compare.P1.peaklat.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.324323 | 0.000005 | -1.624043 | 0.000009 | -1.943702 | 0.000010 |
| small.dark.negVSneut | -0.436943 | 0.000004 | -0.680494 | 0.000003 | -0.963732 | 0.000002 |
| large.bright.negVSneut | -1.288675 | 0.000005 | -1.586022 | 0.000008 | -1.904176 | 0.000009 |
| small.bright.negVSneut | -1.432421 | 0.000005 | -1.739395 | 0.000010 | -2.063643 | 0.000012 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in P1 peak latency:
- large size, high contrast: 0.1971;
- small size, high contrast: 0.506367;
- large size, low contrast: 0.204738;
- small size, low contrast: 0.175627.
Exploratory model selection
for (k in 1:length(scaling.factor)) {
P1.peaklat.anovaBF <- anovaBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "P1" & var == "peak.lat"),
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(P1.peaklat.anovaBF,
file = here::here(paste0("EEG/main/models/P1_peaklat_anovaBF_", scaling.factor[k], ".rds")))
}compare.P1.peaklat.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.P1.peaklat.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load model
P1.peaklat.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/P1_peaklat_anovaBF_", scaling.factor[k], ".rds"))
)
compare.P1.peaklat.anovaBF.temp[, k] <- P1.peaklat.anovaBF@bayesFactor$bf
compare.P1.peaklat.anovaBF.perc.err.temp[, k] <- P1.peaklat.anovaBF@bayesFactor$error * 100
}
compare.P1.peaklat.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.P1.peaklat.anovaBF.temp[, 1], "nar.p.err" = compare.P1.peaklat.anovaBF.perc.err.temp[, 1],
"med" = compare.P1.peaklat.anovaBF.temp[, 2], "med.p.err" = compare.P1.peaklat.anovaBF.perc.err.temp[, 2],
"wid" = compare.P1.peaklat.anovaBF.temp[, 3], "wid.p.err" = compare.P1.peaklat.anovaBF.perc.err.temp[, 3]
)
compare.P1.peaklat.anovaBF <- compare.P1.peaklat.anovaBF[order(compare.P1.peaklat.anovaBF$med, decreasing = TRUE), ]
kable(compare.P1.peaklat.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | emo | 1.83285 | 5.23975 | 2.15809 | 6.90986 | 2.43201 | 6.27148 |
| 7 | size | 1.82995 | 5.12262 | 2.12552 | 6.99069 | 2.47590 | 6.37073 |
| 3 | emo:size | 1.63757 | 4.98234 | 2.00774 | 6.91695 | 2.32132 | 6.39586 |
| 2 | cont:emo | 1.71518 | 5.08172 | 1.97646 | 6.72197 | 2.33141 | 6.17120 |
| 1 | cont:emo:size | 1.24054 | 5.14424 | 1.45722 | 6.87122 | 1.75279 | 6.16118 |
| 4 | cont:size | -9.05984 | 5.82549 | -9.19382 | 6.99118 | -8.97445 | 6.56233 |
| 6 | cont | -11.88366 | 5.75009 | -11.77912 | 7.21225 | -11.63297 | 6.35138 |
Exploratory analyses suggest that omitting emotion from the full model improves fitting by 8.654611 times. Removing size, contrast x emotion, size x emotion, and size x contrast x emotion also improves fitting by 8.377287, 7.446467, 7.217118, and 4.294014 times, respectively.
Conversely, omitting contrast x size lowers the explanatory value of the resulting model by 9836.163801 times. Finally, omitting the factor contrast is maximally detrimental, as it would lower the explanatory value of the resulting model by 1.30499e+05 times.
N1
Peak amplitude
Peak amplitude: negative value larger than the 3 points (~10 ms at 256 Hz sampling rate) on either side of the peak.
Time window: 150-260 ms post-stimulus onset.
Electrode cluster: TP7 P7 P9 TP8 P8 P10.
# summarize data
summary.N1.peakamp <-
summarySEwithin(
subset(data.peaks, component == "N1" & var == "peak.amp"),
"value",
withinvars = c("size", "cont", "emo", "visualfeats"),
idvar = "participant"
)
kable(summary.N1.peakamp)| size | cont | emo | visualfeats | N | value | sd | se | ci |
|---|---|---|---|---|---|---|---|---|
| large | dark | neutral | large dark | 40 | -3.497500 | 1.130311 | 0.178718 | 0.361491 |
| large | dark | negative | large dark | 40 | -3.506750 | 1.046491 | 0.165465 | 0.334684 |
| large | bright | neutral | large bright | 40 | -3.051325 | 0.835515 | 0.132107 | 0.267211 |
| large | bright | negative | large bright | 40 | -3.067950 | 0.711714 | 0.112532 | 0.227617 |
| small | dark | neutral | small dark | 40 | -3.302025 | 1.089445 | 0.172256 | 0.348421 |
| small | dark | negative | small dark | 40 | -3.243100 | 0.881943 | 0.139447 | 0.282059 |
| small | bright | neutral | small bright | 40 | -1.244675 | 1.309893 | 0.207112 | 0.418924 |
| small | bright | negative | small bright | 40 | -0.992025 | 1.719284 | 0.271843 | 0.549854 |
# pirateplot
pirateplot(
formula = value ~ emo + visualfeats,
data = subset(data.peaks, component == "N1" & var == "peak.amp"),
main = "N1 peak amplitude",
xlim = NULL,
xlab = "",
ylim = c(-10, 3),
ylab = expression(paste("amplitude (", mu, "V)")),
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
N1.peakamp.sizeplusemo.BF <- lmBF(
value ~ size + emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.sizeplusemo.BF, # save model
file = here::here(paste0("EEG/main/models/N1_peakamp_sizeplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size and emotion
N1.peakamp.sizebyemo.BF <- lmBF(
value ~ size * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/N1_peakamp_sizebyemo_BF_", scaling.factor[k], ".rds")))
### main effects of contrast and emotion
N1.peakamp.contplusemo.BF <- lmBF(
value ~ cont + emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.contplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_peakamp_contplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of contrast and emotion
N1.peakamp.contbyemo.BF <- lmBF(
value ~ cont * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.contbyemo.BF,
file = here::here(paste0("EEG/main/models/N1_peakamp_contbyemo_BF_", scaling.factor[k], ".rds")))
### main effects of size, contrast, and emotion
N1.peakamp.sizepluscontplusemo.BF <- lmBF(
value ~ size + cont + emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_peakamp_sizepluscontplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size, contrast, and emotion
N1.peakamp.sizebycontbyemo.BF <- lmBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/N1_peakamp_sizebycontbyemo_BF_", scaling.factor[k], ".rds")))
}compare.N1.peakamp.BF <- matrix(NA, 6, length(scaling.factor))
compare.N1.peakamp.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
N1.peakamp.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.peakamp.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
N1.peakamp.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_contplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.peakamp.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_contbyemo_BF_", scaling.factor[k], ".rds"))
)
N1.peakamp.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.peakamp.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
compare.N1.peakamp.BF[, k] <- c(
N1.peakamp.sizeplusemo.BF@bayesFactor$bf,
N1.peakamp.sizebyemo.BF@bayesFactor$bf,
N1.peakamp.contplusemo.BF@bayesFactor$bf,
N1.peakamp.contbyemo.BF@bayesFactor$bf,
N1.peakamp.sizepluscontplusemo.BF@bayesFactor$bf,
N1.peakamp.sizebycontbyemo.BF@bayesFactor$bf
)
compare.N1.peakamp.perc.err[, k] <- c(
N1.peakamp.sizeplusemo.BF@bayesFactor$error * 100,
N1.peakamp.sizebyemo.BF@bayesFactor$error * 100,
N1.peakamp.contplusemo.BF@bayesFactor$error * 100,
N1.peakamp.contbyemo.BF@bayesFactor$error * 100,
N1.peakamp.sizepluscontplusemo.BF@bayesFactor$error * 100,
N1.peakamp.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
compare.N1.peakamp <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.N1.peakamp.BF[, 1], "nar.p.err" = compare.N1.peakamp.perc.err[, 1],
"med" = compare.N1.peakamp.BF[, 2], "med.p.err" = compare.N1.peakamp.perc.err[, 2],
"wid" = compare.N1.peakamp.BF[, 3], "wid.p.err" = compare.N1.peakamp.perc.err[, 3]
)
compare.N1.peakamp <- compare.N1.peakamp[order(compare.N1.peakamp$med, decreasing = TRUE), ]
kable(compare.N1.peakamp)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 6 | size x cont x emo | 132.5116 | 2.188652 | 131.3568 | 2.215321 | 129.8644 | 2.308238 |
| 5 | size + cont + emo | 120.4792 | 1.014891 | 120.2075 | 1.004733 | 119.7191 | 1.040088 |
| 3 | contr + emo | 96.7546 | 0.868479 | 96.4366 | 0.873540 | 96.0086 | 0.889878 |
| 4 | cont x emo | 95.0109 | 1.083816 | 94.4108 | 1.126364 | 93.6376 | 1.141639 |
| 1 | size + emo | 86.8748 | 0.852969 | 86.4728 | 0.874806 | 85.9936 | 0.916449 |
| 2 | size x emo | 85.2270 | 1.087511 | 84.5178 | 1.114712 | 83.7137 | 1.168779 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size x cont x emo ought to be preferred.
The best model (size x cont x emo) explains the observed data 6.95162e+04 times better than the second best model (size + cont + emo).
Paired comparisons
N1.peakamp.posthocBF <-
subset(data.peaks, component == "N1" & var == "peak.amp") %>%
select(participant, condition, value) %>%
split(.$condition)
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
N1.peakamp.posthocBF.large.dark.negVSneut <- ttestBF(
N1.peakamp.posthocBF$`negativeLargeDark`$value,
N1.peakamp.posthocBF$`neutralLargeDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peakamp.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds")))
### small, dark, negative vs. neutral
N1.peakamp.posthocBF.small.dark.negVSneut <- ttestBF(
N1.peakamp.posthocBF$`negativeSmallDark`$value,
N1.peakamp.posthocBF$`neutralSmallDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peakamp.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds")))
### large, bright, negative vs. neutral
N1.peakamp.posthocBF.large.bright.negVSneut <- ttestBF(
N1.peakamp.posthocBF$`negativeLargeBright`$value,
N1.peakamp.posthocBF$`neutralLargeBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peakamp.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds")))
### small, bright, negative vs. neutral
N1.peakamp.posthocBF.small.bright.negVSneut <- ttestBF(
N1.peakamp.posthocBF$`negativeSmallBright`$value,
N1.peakamp.posthocBF$`neutralSmallBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peakamp.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds")))
}compare.N1.peakamp.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.N1.peakamp.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
N1.peakamp.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
N1.peakamp.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
N1.peakamp.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
N1.peakamp.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### model comparison
compare.N1.peakamp.posthocBF[, k] <- c(
N1.peakamp.posthocBF.large.dark.negVSneut@bayesFactor$bf,
N1.peakamp.posthocBF.small.dark.negVSneut@bayesFactor$bf,
N1.peakamp.posthocBF.large.bright.negVSneut@bayesFactor$bf,
N1.peakamp.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
# percentage of error
compare.N1.peakamp.posthocBF.perc.err[, k] <- c(
N1.peakamp.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
N1.peakamp.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
N1.peakamp.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
N1.peakamp.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
# summary
compare.N1.peakamp.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.N1.peakamp.posthocBF[, 1], "nar.p.err" = compare.N1.peakamp.posthocBF.perc.err[, 1],
"med" = compare.N1.peakamp.posthocBF[, 2], "med.p.err" = compare.N1.peakamp.posthocBF.perc.err[, 2],
"wid" = compare.N1.peakamp.posthocBF[, 3], "wid.p.err" = compare.N1.peakamp.posthocBF.perc.err[, 3]
)
kable(compare.N1.peakamp.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.45681 | 0.000005 | -1.76543 | 0.000010 | -2.09072 | 0.000012 |
| small.dark.negVSneut | -1.39664 | 0.000005 | -1.70120 | 0.000009 | -2.02393 | 0.000011 |
| large.bright.negVSneut | -1.45587 | 0.000005 | -1.76442 | 0.000010 | -2.08967 | 0.000012 |
| small.bright.negVSneut | -1.05302 | 0.000005 | -1.33493 | 0.000006 | -1.64322 | 0.000007 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in N1 peak amplitude:
- large size, high contrast: 0.171113;
- small size, high contrast: 0.182464;
- large size, low contrast: 0.171285;
- small size, low contrast: 0.263177.
Exploratory model selection
for (k in 1:length(scaling.factor)) {
N1.peakamp.anovaBF <- anovaBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.amp"),
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(N1.peakamp.anovaBF,
file = here::here(paste0("EEG/main/models/N1_peakamp_anovaBF_", scaling.factor[k], ".rds")))
}compare.N1.peakamp.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.N1.peakamp.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load model
N1.peakamp.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/N1_peakamp_anovaBF_", scaling.factor[k], ".rds"))
)
compare.N1.peakamp.anovaBF.temp[, k] <- N1.peakamp.anovaBF@bayesFactor$bf
compare.N1.peakamp.anovaBF.perc.err.temp[, k] <- N1.peakamp.anovaBF@bayesFactor$error * 100
}
compare.N1.peakamp.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.N1.peakamp.anovaBF.temp[, 1], "nar.p.err" = compare.N1.peakamp.anovaBF.perc.err.temp[, 1],
"med" = compare.N1.peakamp.anovaBF.temp[, 2], "med.p.err" = compare.N1.peakamp.anovaBF.perc.err.temp[, 2],
"wid" = compare.N1.peakamp.anovaBF.temp[, 3], "wid.p.err" = compare.N1.peakamp.anovaBF.perc.err.temp[, 3]
)
compare.N1.peakamp.anovaBF <- compare.N1.peakamp.anovaBF[order(compare.N1.peakamp.anovaBF$med, decreasing = TRUE), ]
kable(compare.N1.peakamp.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | emo | 1.99368 | 2.89754 | 2.31218 | 2.64091 | 2.60994 | 3.10225 |
| 2 | cont:emo | 1.75228 | 2.87886 | 2.02680 | 2.66983 | 2.35592 | 3.10442 |
| 3 | emo:size | 1.59916 | 2.93720 | 1.93184 | 2.67864 | 2.20840 | 3.10644 |
| 1 | cont:emo:size | 1.43732 | 2.91708 | 1.70554 | 2.67589 | 2.01439 | 3.11263 |
| 4 | cont:size | -16.66927 | 2.97567 | -16.80553 | 2.81724 | -16.89753 | 3.25349 |
| 7 | size | -27.15137 | 2.94878 | -27.29704 | 2.77292 | -27.30376 | 3.26466 |
| 6 | cont | -38.17203 | 2.95101 | -38.39241 | 2.73704 | -38.46348 | 3.23530 |
Exploratory analyses suggest that omitting emotion from the full model improves fitting by 10.096435 times. Removing contrast x emotion, size x emotion, and size x contrast x emotion also improves fitting by 7.589796, 6.902212, and 5.504349 times, respectively.
Conversely, omitting contrast x size or size lowers the explanatory value of the resulting model by 1.98861e+07 and 7.16065e+11 times, respectively. Finally, omitting the factor contrast is maximally detrimental, as it would lower the explanatory value of the resulting model by 4.71643e+16 times.
Peak latency
Peak latency: time (in ms) of the negative value larger than the 3 points (~10 ms at 256 Hz sampling rate) on either side of the peak.
Time window: 150-260 ms post-stimulus onset.
Electrode cluster: TP7 P7 P9 TP8 P8 P10.
summary.N1.peaklat <-
summarySEwithin(
subset(data.peaks, component == "N1" & var == "peak.lat"),
"value",
withinvars = c("size", "cont", "emo", "visualfeats"),
idvar = "participant"
)
kable(summary.N1.peaklat)| size | cont | emo | visualfeats | N | value | sd | se | ci |
|---|---|---|---|---|---|---|---|---|
| large | dark | neutral | large dark | 40 | 187.793 | 19.4473 | 3.07488 | 6.21953 |
| large | dark | negative | large dark | 40 | 190.235 | 23.3587 | 3.69333 | 7.47047 |
| large | bright | neutral | large bright | 40 | 202.149 | 20.1355 | 3.18370 | 6.43963 |
| large | bright | negative | large bright | 40 | 210.742 | 17.4317 | 2.75619 | 5.57492 |
| small | dark | neutral | small dark | 40 | 199.316 | 21.2240 | 3.35581 | 6.78777 |
| small | dark | negative | small dark | 40 | 196.484 | 17.4944 | 2.76611 | 5.59498 |
| small | bright | neutral | small bright | 40 | 214.942 | 30.5002 | 4.82250 | 9.75444 |
| small | bright | negative | small bright | 40 | 213.770 | 32.1088 | 5.07684 | 10.26888 |
# pirateplot
pirateplot(
formula = value ~ emo + visualfeats,
data = subset(data.peaks, component == "N1" & var == "peak.lat"),
main = "N1 peak latency",
xlim = NULL,
xlab = "",
ylim = c(150, 260),
ylab = "time (ms)",
inf.method = "hdi",
inf.within = "participant",
hdi.iter = 5000,
cap.beans = TRUE,
point.cex = 1.3,
pal = pirateplot.palette
)
for (k in 1:length(scaling.factor)) {
### main effects of size and emotion
N1.peaklat.sizeplusemo.BF <- lmBF(
value ~ size + emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.sizeplusemo.BF, # save model
file = here::here(paste0("EEG/main/models/N1_peaklat_sizeplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size and emotion
N1.peaklat.sizebyemo.BF <- lmBF(
value ~ size * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.sizebyemo.BF,
file = here::here(paste0("EEG/main/models/N1_peaklat_sizebyemo_BF_", scaling.factor[k], ".rds")))
### main effects of contrast and emotion
N1.peaklat.contplusemo.BF <- lmBF(
value ~ cont + emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.contplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_peaklat_contplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of contrast and emotion
N1.peaklat.contbyemo.BF <- lmBF(
value ~ cont * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.contbyemo.BF,
file = here::here(paste0("EEG/main/models/N1_peaklat_contbyemo_BF_", scaling.factor[k], ".rds")))
### main effects of size, contrast, and emotion
N1.peaklat.sizepluscontplusemo.BF <- lmBF(
value ~ size + cont + emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.sizepluscontplusemo.BF,
file = here::here(paste0("EEG/main/models/N1_peaklat_sizepluscontplusemo_BF_", scaling.factor[k], ".rds")))
### interactive effects of size, contrast, and emotion
N1.peaklat.sizebycontbyemo.BF <- lmBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
rscaleCont = "medium",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.sizebycontbyemo.BF,
file = here::here(paste0("EEG/main/models/N1_peaklat_sizebycontbyemo_BF_", scaling.factor[k], ".rds")))
}compare.N1.peaklat.BF <- matrix(NA, 6, length(scaling.factor))
compare.N1.peaklat.perc.err <- matrix(NA, 6, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
N1.peaklat.sizeplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_sizeplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.peaklat.sizebyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_sizebyemo_BF_", scaling.factor[k], ".rds"))
)
N1.peaklat.contplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_contplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.peaklat.contbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_contbyemo_BF_", scaling.factor[k], ".rds"))
)
N1.peaklat.sizepluscontplusemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_sizepluscontplusemo_BF_", scaling.factor[k], ".rds"))
)
N1.peaklat.sizebycontbyemo.BF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_sizebycontbyemo_BF_", scaling.factor[k], ".rds"))
)
compare.N1.peaklat.BF[, k] <- c(
N1.peaklat.sizeplusemo.BF@bayesFactor$bf,
N1.peaklat.sizebyemo.BF@bayesFactor$bf,
N1.peaklat.contplusemo.BF@bayesFactor$bf,
N1.peaklat.contbyemo.BF@bayesFactor$bf,
N1.peaklat.sizepluscontplusemo.BF@bayesFactor$bf,
N1.peaklat.sizebycontbyemo.BF@bayesFactor$bf
)
compare.N1.peaklat.perc.err[, k] <- c(
N1.peaklat.sizeplusemo.BF@bayesFactor$error * 100,
N1.peaklat.sizebyemo.BF@bayesFactor$error * 100,
N1.peaklat.contplusemo.BF@bayesFactor$error * 100,
N1.peaklat.contbyemo.BF@bayesFactor$error * 100,
N1.peaklat.sizepluscontplusemo.BF@bayesFactor$error * 100,
N1.peaklat.sizebycontbyemo.BF@bayesFactor$error * 100
)
}
compare.N1.peaklat <- data.frame(
"model" = c("size + emo", "size x emo", "contr + emo", "cont x emo", "size + cont + emo", "size x cont x emo"),
"nar" = compare.N1.peaklat.BF[, 1], "nar.p.err" = compare.N1.peaklat.perc.err[, 1],
"med" = compare.N1.peaklat.BF[, 2], "med.p.err" = compare.N1.peaklat.perc.err[, 2],
"wid" = compare.N1.peaklat.BF[, 3], "wid.p.err" = compare.N1.peaklat.perc.err[, 3]
)
compare.N1.peaklat <- compare.N1.peaklat[order(compare.N1.peaklat$med, decreasing = TRUE), ]
kable(compare.N1.peaklat)| model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | size + cont + emo | 26.6594 | 1.16392 | 26.01855 | 1.24868 | 25.21809 | 1.33281 |
| 3 | contr + emo | 24.2288 | 1.01534 | 23.82961 | 1.04759 | 23.31993 | 1.08492 |
| 4 | cont x emo | 22.7133 | 1.26208 | 21.98327 | 1.30426 | 21.12926 | 1.37474 |
| 6 | size x cont x emo | 21.0690 | 2.39584 | 19.23063 | 2.65336 | 17.18204 | 3.00888 |
| 1 | size + emo | 11.1017 | 1.15394 | 10.51553 | 1.20461 | 9.87124 | 1.21072 |
| 2 | size x emo | 10.1246 | 1.40983 | 9.21805 | 1.47848 | 8.28491 | 1.51659 |
When using a JZS prior with scaling factor r = 0.707 placed on standardized effect sizes, the model size + cont + emo ought to be preferred.
The best model (size + cont + emo) explains the observed data 8.925722 times better than the second best model (contr + emo).
Paired comparisons
N1.peaklat.posthocBF <-
subset(data.peaks, component == "N1" & var == "peak.lat") %>%
select(participant, condition, value) %>%
split(.$condition)
for (k in 1:length(scaling.factor)) {
### large, dark, negative vs. neutral
N1.peaklat.posthocBF.large.dark.negVSneut <- ttestBF(
N1.peaklat.posthocBF$`negativeLargeDark`$value,
N1.peaklat.posthocBF$`neutralLargeDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peaklat.posthocBF.large.dark.negVSneut, # save model
file = here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds")))
### small, dark, negative vs. neutral
N1.peaklat.posthocBF.small.dark.negVSneut <- ttestBF(
N1.peaklat.posthocBF$`negativeSmallDark`$value,
N1.peaklat.posthocBF$`neutralSmallDark`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peaklat.posthocBF.small.dark.negVSneut,
file = here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds")))
### large, bright, negative vs. neutral
N1.peaklat.posthocBF.large.bright.negVSneut <- ttestBF(
N1.peaklat.posthocBF$`negativeLargeBright`$value,
N1.peaklat.posthocBF$`neutralLargeBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peaklat.posthocBF.large.bright.negVSneut,
file = here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds")))
### small, bright, negative vs. neutral
N1.peaklat.posthocBF.small.bright.negVSneut <- ttestBF(
N1.peaklat.posthocBF$`negativeSmallBright`$value,
N1.peaklat.posthocBF$`neutralSmallBright`$value,
mu = 0,
paired = TRUE,
method = "simple",
iterations = niter,
rscale = scaling.factor[k]
)
saveRDS(N1.peaklat.posthocBF.small.bright.negVSneut,
file = here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds")))
}compare.N1.peaklat.posthocBF <- matrix(NA, 4, length(scaling.factor))
compare.N1.peaklat.posthocBF.perc.err <- matrix(NA, 4, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load models
N1.peaklat.posthocBF.large.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_large_dark_negVSneut_", scaling.factor[k], ".rds"))
)
N1.peaklat.posthocBF.small.dark.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_small_dark_negVSneut_", scaling.factor[k], ".rds"))
)
N1.peaklat.posthocBF.large.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_large_bright_negVSneut_", scaling.factor[k], ".rds"))
)
N1.peaklat.posthocBF.small.bright.negVSneut <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_posthocBF_small_bright_negVSneut_", scaling.factor[k], ".rds"))
)
### model comparison
compare.N1.peaklat.posthocBF[, k] <- c(
N1.peaklat.posthocBF.large.dark.negVSneut@bayesFactor$bf,
N1.peaklat.posthocBF.small.dark.negVSneut@bayesFactor$bf,
N1.peaklat.posthocBF.large.bright.negVSneut@bayesFactor$bf,
N1.peaklat.posthocBF.small.bright.negVSneut@bayesFactor$bf
)
# percentage of error
compare.N1.peaklat.posthocBF.perc.err[, k] <- c(
N1.peaklat.posthocBF.large.dark.negVSneut@bayesFactor$error * 100,
N1.peaklat.posthocBF.small.dark.negVSneut@bayesFactor$error * 100,
N1.peaklat.posthocBF.large.bright.negVSneut@bayesFactor$error * 100,
N1.peaklat.posthocBF.small.bright.negVSneut@bayesFactor$error * 100
)
}
# summary
compare.N1.peaklat.posthocBF <- data.frame(
"comparison" = c("large.dark.negVSneut", "small.dark.negVSneut", "large.bright.negVSneut", "small.bright.negVSneut"),
"nar" = compare.N1.peaklat.posthocBF[, 1], "nar.p.err" = compare.N1.peaklat.posthocBF.perc.err[, 1],
"med" = compare.N1.peaklat.posthocBF[, 2], "med.p.err" = compare.N1.peaklat.posthocBF.perc.err[, 2],
"wid" = compare.N1.peaklat.posthocBF[, 3], "wid.p.err" = compare.N1.peaklat.posthocBF.perc.err[, 3]
)
kable(compare.N1.peaklat.posthocBF)| comparison | nar | nar.p.err | med | med.p.err | wid | wid.p.err |
|---|---|---|---|---|---|---|
| large.dark.negVSneut | -1.291370 | 0.000005 | -1.588896 | 0.000008 | -1.907163 | 0.000009 |
| small.dark.negVSneut | -1.210616 | 0.000005 | -1.502803 | 0.000008 | -1.817672 | 0.000008 |
| large.bright.negVSneut | 0.503694 | 0.000002 | 0.313551 | 0.000001 | 0.066594 | 0.000000 |
| small.bright.negVSneut | -1.445018 | 0.000005 | -1.752844 | 0.000010 | -2.077628 | 0.000012 |
Follow-up contrasts reveal that the effect of emotion does not explain changes in N1 peak latency:
- large size, high contrast: 0.204151;
- small size, high contrast: 0.222506;
- large size, low contrast: 1.368275;
- small size, low contrast: 0.173281.
Exploratory model selection
for (k in 1:length(scaling.factor)) {
N1.peaklat.anovaBF <- anovaBF(
value ~ size * cont * emo + participant,
subset(data.peaks, component == "N1" & var == "peak.lat"),
whichModels = "top",
whichRandom = "participant",
rscaleFixed = scaling.factor[k],
rscaleRandom = "nuisance",
method = "simple",
iterations = niter
)
saveRDS(N1.peaklat.anovaBF,
file = here::here(paste0("EEG/main/models/N1_peaklat_anovaBF_", scaling.factor[k], ".rds")))
}compare.N1.peaklat.anovaBF.temp <- matrix(NA, 7, length(scaling.factor))
compare.N1.peaklat.anovaBF.perc.err.temp <- matrix(NA, 7, length(scaling.factor))
### model comparison across scaling factors
for (k in 1:length(scaling.factor)) {
### load model
N1.peaklat.anovaBF <- readRDS(
here::here(paste0("EEG/main/models/N1_peaklat_anovaBF_", scaling.factor[k], ".rds"))
)
compare.N1.peaklat.anovaBF.temp[, k] <- N1.peaklat.anovaBF@bayesFactor$bf
compare.N1.peaklat.anovaBF.perc.err.temp[, k] <- N1.peaklat.anovaBF@bayesFactor$error * 100
}
compare.N1.peaklat.anovaBF <- data.frame(
"omit from full model" = c("cont:emo:size", "cont:emo", "emo:size", "cont:size", "emo", "cont", "size"),
"nar" = compare.N1.peaklat.anovaBF.temp[, 1], "nar.p.err" = compare.N1.peaklat.anovaBF.perc.err.temp[, 1],
"med" = compare.N1.peaklat.anovaBF.temp[, 2], "med.p.err" = compare.N1.peaklat.anovaBF.perc.err.temp[, 2],
"wid" = compare.N1.peaklat.anovaBF.temp[, 3], "wid.p.err" = compare.N1.peaklat.anovaBF.perc.err.temp[, 3]
)
compare.N1.peaklat.anovaBF <- compare.N1.peaklat.anovaBF[order(compare.N1.peaklat.anovaBF$med, decreasing = TRUE), ]
kable(compare.N1.peaklat.anovaBF)| omit.from.full.model | nar | nar.p.err | med | med.p.err | wid | wid.p.err | |
|---|---|---|---|---|---|---|---|
| 5 | emo | 1.844583 | 3.24770 | 2.26548 | 3.67449 | 2.56387 | 3.76734 |
| 4 | cont:size | 1.761787 | 3.32271 | 2.05090 | 3.58470 | 2.43566 | 3.84750 |
| 2 | cont:emo | 1.467660 | 3.28640 | 1.84285 | 3.57786 | 2.14626 | 3.79942 |
| 1 | cont:emo:size | 1.303180 | 3.25963 | 1.70487 | 3.64869 | 1.99009 | 3.91498 |
| 3 | emo:size | 0.864005 | 3.37079 | 1.21360 | 3.76316 | 1.52784 | 3.96675 |
| 7 | size | -2.447698 | 3.42704 | -2.20890 | 3.66095 | -1.89773 | 4.08988 |
| 6 | cont | -15.750420 | 3.56978 | -15.66128 | 3.82710 | -15.40421 | 4.17324 |
Exploratory analyses suggest that omitting emotion from the full model improves fitting by 9.6358 times. Removing size x contrast, contrast x emotion, size x contrast x emotion, and size x emotion also improves fitting by 7.774863, 6.314525, 5.50065, and 3.365584 times, respectively.
Conversely, omitting size lowers the explanatory value of the resulting model by 9.105681 times. Finally, omitting the factor contrast is maximally detrimental, as it would lower the explanatory value of the resulting model by 6.33296e+06 times.
Interpolated Channels
elecs <-
read_csv(here::here("EEG/main/data/electrodes/elecs.csv")) %>%
mutate(
SSJ = as.factor(SSJ),
ELECS = as.factor(ELECS),
P1 = case_when(
ELECS == "P7" ~ 1,
ELECS == "P9" ~ 1,
ELECS == "PO7" ~ 1,
ELECS == "O1" ~ 1,
ELECS == "O2" ~ 1,
ELECS == "PO8" ~ 1,
ELECS == "P8" ~ 1,
ELECS == "P10" ~ 1,
TRUE ~ 0
),
N1 = case_when(
ELECS == "TP7" ~ 1,
ELECS == "P7" ~ 1,
ELECS == "P9" ~ 1,
ELECS == "TP8" ~ 1,
ELECS == "P8" ~ 1,
ELECS == "P10" ~ 1,
TRUE ~ 0
),
EPN = case_when(
ELECS == "P9" ~ 1,
ELECS == "PO7" ~ 1,
ELECS == "PO3" ~ 1,
ELECS == "O1" ~ 1,
ELECS == "Oz" ~ 1,
ELECS == "Iz" ~ 1,
ELECS == "O2" ~ 1,
ELECS == "PO4" ~ 1,
ELECS == "PO8" ~ 1,
ELECS == "P8" ~ 1,
ELECS == "P10" ~ 1,
TRUE ~ 0
),
LPP = case_when(
ELECS == "P1" ~ 1,
ELECS == "Pz" ~ 1,
ELECS == "P2" ~ 1,
ELECS == "P4" ~ 1,
ELECS == "P6" ~ 1,
ELECS == "P8" ~ 1,
ELECS == "P10" ~ 1,
ELECS == "POz" ~ 1,
ELECS == "PO4" ~ 1,
ELECS == "PO8" ~ 1,
TRUE ~ 0
)
)
interp.chans <-
elecs %>%
group_by(SSJ) %>%
dplyr::summarize(
tot.interp = n(),
P1.interp = sum(P1),
N1.interp = sum(N1),
EPN.interp = sum(EPN),
LPP.interp = sum(LPP)
)
descr.interp.chans <-
interp.chans %>%
summarize(
P1.interp.min = min(P1.interp),
P1.interp.max = max(P1.interp),
N1.interp.min = min(N1.interp),
N1.interp.max = max(N1.interp),
EPN.interp.min = min(EPN.interp),
EPN.interp.max = max(EPN.interp),
LPP.interp.min = min(LPP.interp),
LPP.interp.max = max(LPP.interp)
) %>%
print()## P1.interp.min P1.interp.max N1.interp.min N1.interp.max EPN.interp.min EPN.interp.max LPP.interp.min LPP.interp.max
## 1 0 1 0 2 0 2 0 2The interpolated channels were mostly identifed outside of the clusters selected for the ERP components defnition (max interpolated channels in clusters: 2); therefore, any potential distortions of the EEG signal due to interpolation was negligible.